Raid5 is probably the best compromise between increased speed over single drive vs. data reliability.
...
Raid5 - read/write speed increased by factor equal to number of drives minus 1. any of the drives can fail and all data is retained. (if 2 drives fail simultaneously, then all data is lost)
RAID5 can be slow at writing if not ideally optimised for the filesystem using it. A number of them get around this problem by merging the two in some way, such as Sun's ZFS and the Veritas storage system. I prefer RAID6 as it has two XOR devices per array, though that's probably getting a little paranoid.
RAID is for higher speed or greater uptime (or both). It really has no place outside of production servers and is absolutely NOT a backup solution. In many ways, RAID makes data more volatile, and many people consider it a "backup replacement" when nothing could be further from the truth.
I hope nobody assumed I was suggesting that RAID is a substitute for backups, as it absolutely is not, as you rightly point out. The only substitute for a backup system is another backup system; take a worst-case situation where a big enough power surge fries your computer, no RAID system is going to recover if all its HDDs are dead. Write caching should definitely be disabled using RAID (in fact IMHO it should never be enabled unless the data really is considered disposable) though with the various types of command queuing available on SCSI and SATA storage, there should be little performance penalty (if any) by disabling it anyway.
I would strongly disagree that its only place is in production servers, though: few if any of us backup as regularly as we probably should, and I have experienced both the data-loss and hassle of recovering a small server on more than one occasion. I don't intend to do it again if I can possibly avoid it.