Overall, I agree with the OP's assessment of the strengths and weaknesses of the various systems. As stated, a "partial" leveling approach with regional variations and occasional static placements appears to thoroughly beat either an "all static" or an "all leveled and scaled" one.
Yes, each map area in the game can have several virtual height maps that define the minimum and maximum level for several factors in those maps.
Each height map is created by plasma a fractal formula that is a normal way to make height maps, and is used in several landscape generation programs, but here we want to use those height maps for other reasons.
Those formulas let us define the exact height of several places, and then the engine fills the other areas with semi-random connected height values in a way that the predefined height values are fitted perfectly to the surface height of the whole map, so if a point is near one of the predefined points, then it's height value is near the height value of the predefined point, but it might be a bit lower or upper, and different predefined points would affect their surrounding height map a bit, but the most important factor is the random seed of the whole map.
This way, we can have two virtual height maps that define the minimum and maximum possible level of monsters in each area of the game world, and we can have two virtual height maps that define minimum and maximum level of the loot found on those monsters, or items found on shops and so on...
We can have the height maps that show the player character's fame and infamy in the game world, or the influence of different guilds and faction over the game world area, or the density of different character races over the landscape to aid the automatic spawners of random population, or the different monster type density to define areas that spawn animal more than daedra and so on...
Those virtual height map that define the minimum and the maximum of possible level of spawned monsters, are affected by player level, so if the player level did not affect the spawned monster level in a place, then the level of the monster would be defined by a probability of a normal bell shape within those limits, but if the effect of the player level was added, then the bell shape would
lean toward the player level, so if the player level is below the middle of the maximum and minimum level value of the place, then the probability bell shape would lean to lower values, and if the player level is above the middle point, then the bell shape would lean toward higher values.
This would affect the probable level of monsters in an area, and the monsters and items could be picked from random lists that are affected by the difficulty level calculated from this formula.
The great feature of those virtual height maps are that they can be changed in the real time as a script can define or change the value of a point on the map and the surrounding height map would change shape to embrace the change of the value in that point, so if a battle happens in an area, then a script can raise the maximum difficulty level of a point on the middle of that place, and the whole area would have a higher danger level, and after the battle, the script can remove the change on that point and revert the area to the previous state, and so on...
Or if the player does a good deed for a faction in an area, then a script can raise the virtual height map representing his reputation in the faction on that place and it would affect the surrounding area as well, and this formula can be used in a lot of different aspects of the game.
The idea of adjusting skill gain based on the difficulty of the task is a noteworthy addition, but should encompass more than just combat opponents: casting a high-level spell with less than a 100% chance of success should teach you more than rehashing the same old "training spell" for the 600th time. Skill gain rates should also be affected by their selection as Major, Minor, etc. It should be easier to increase a Major skill, but raising it should be no more beneficial to Attributes than any other skill.
Yes, each usage of any skill, should raise the player character's experience in that skill by a defined value that depends on the difficulty of the deed done, so a character with low skill of lock picking can pick a hard lock then he should have a good boost in his lock picking experience points, but if a skilled thief picks an easy lock, then it would not affect his lock picking experience that much.
This trend should be for all the skill usages, so for instance we should have potions that are easy to make and elixirs that are hard to make and those affect the alchemy skill differently, and so on...
I'd prefer "perks" to be "phased in", where you "learn" the perk, but have to "build" it by taking it multiple times to make it really effective. For example, Resistance to Magic could begin with your character gaining an initial 5% reduction in magical-based damage and negative effects, which is trivial. By re-selecting that perk each level, your resistance would slowly grow to 10%, 15%, etc. Either you would have a lot of separate trivial "tricks" that wouldn't be all that effective or you'd specialize your character and develop one or two "gimmicks" into truly remarkable abilities. The same could go for spells, where you'd learn the basic spell, but have to gradually "build" it over time and practice, with occasional paid training or quest rewards to speed the process. MW's spell failure idea was good in principle, but awkward in implementation. Allowing you to manually adjust the strength/duration/area of spells to either maximize results or minimize risk of failure would be much better, since failure would be controllable. OB's "automatic perks" that opened up the next level of spells or abilities at each 25 point increment was far too contrived and artificial for my liking, as well as taking away the entire concept of "risk versus reward", because there were no risks.
If the active perks (or spells) have effects that depend of the level of their respective skill, then those gradual raise of their effectiveness would show as the character gets better in that skill, so for instance each damage dealing spell of a school of magic deals a damage like this:
Fire Bolt: Damage=(1d3)+10sk[10-30]
Which means, if your destruction magic is 10 or below then your "Fire Bolt" spell would deal 1d3 or (1-3) damage, and if your destruction magic is 30 or above then your "Fire Bolt" would deal 1d3+10 or (11-13) damage, and if your destruction magic is 25 then your "Fire Bolt" would deal 1d3+(25-10)*10/(30-10) = 1d3+15*10/20 = 1d3+7 = (8-10).
Fire Ball: Damage=(2d6)+18sk[20-50], Radius=1+3sk[20-50]
Which means, if your destruction magic is 20 or below then your "Fire Ball" spell would deal 2d6 or (2-12) damage, and if your destruction magic is 50 or above then your "Fire Ball" would deal 2d6+18 or (20-30) damage, and if your destruction magic is 25 then your "Fire Ball" would deal 2d6+(25-20)*18/(50-20) = 2d6+5*18/30 = 2d6+3 = (5-15).
And the radius of area effect of "Fire Ball" would be 1ft at skill level of 20 or below and 4ft at level 50 or above and somewhere in between when skill level is between 20 to 50.
The concept of learning by doing is an inherent part of the TES system, and is one of its strong points. Tying Attribute gain to skill gain directly (GCD for MW, nGCD for OB) makes a lot more sense and is far less annoying than using a level-up screen and easily exploitable "multipliers", but reduces the freedom of the player to "shape" their character as they want to. MADD Leveler for MW uses a similar system as GCD/nGCD for most of your attribute gains, but still allows you to manually adjust three attributes by ONE point each. which allows you to "bend" the character's growth a little more toward your preferences (sadly, the mod also causes a small but noticeable increase in level-up rate). A hybrid system seems like a better idea than a single straight-forward approach.
Yes, I support this idea and my Oblivion mod, "Pure Immersion" acted exactly like that, so your attributes raised dynamically as your skill improved, and in the level-up sessions you could raise three of your attributes by just one point.
You could set a difficulty value that defined the amount of attribute raise as your skills advanced, so in lower difficulty value, your attributes raised faster, and in higher difficulty values, your attributes raised slower.
Here I suggested that your attributes do not raise with skill advancement, or their growth should be really slow, but in level-up sessions you could raise one, or two attributes by one point, thus your attributes are really hard to change in the course of the game, and they would define the rate that you could raise the skills that they govern, and also define how much you could get promotion in their governed skills and raise their soft caps.
Condition of weapons and armor, and the cost and ability to repair them, could be a useful limiting factor, as in FO3. By using "repair parts" for the specific material, instead of self-destructing repair hammers, it would make maintenance of exotic and expensive items all but unaffordable for a low-level character. Sure, you can steal, loot, or buy that set of glass armor at L1 (even in good condition), but what are you going to do when it starts to wear down? The repairs (and even the spare parts to do it yourself, IF you have the skill) will cost you more than you can afford, and it's not worth a lot to sell in its pathetic condition.
Yes, that's a great idea, but I think we should make it a bit easier, and let people repair items with their similar items for a full effect, or with their parts and ingredients for a smaller effect, like repairing adamantum shields with adamantum ores, and so on...
And they should be able to disintegrate adamantum boots into adamantum ores in front of the forges and the like.
Most importantly, the differences between a L1 and a L10 character are extreme under the current system, and by L30 there's not a lot in MW that could even make your character break a sweat, while an OB character would either be "god-like" or "wimpy" by that point, depending on how efficiently they leveled. I'd rather see skills "soft capped" at some point to make it exponentially harder to advance above a certain point, depending on whether they're Major, Minor, or Misc, and only see slight Attribute increases over the course of play. Limiting Attribute increases to only 3-5 TOTAL per level (one of them manually selected at level-up, the rest as a direct result of skill usage) would reduce the differences between a low and high level character, and make the game more "playable" for different styles of play, without worrying as much about "efficient leveling". Skills should be the main factor in deciding outcomes, with Attributes secondary and "somewhat tweak able" over time, which would make for a REAL difference in the various races and types, without totally ruling out the possibility of developing a "charismatic Orc" character. You SHOULD be able to do anything in the game world, in theory, but some things should be MUCH harder for some characters than others.
Just like I said above, I support this idea, we should be able to do anything, but some goals should be relatively easier than other goals, so if you want to reach a specific goal, you should generate a proper character for that goal, to make life easier for yourself.