I hate to call everything a driver bug, but multisampling is only supposed to run the pixel shader once and multisample polygon edges only. It shouldn't be fetching neon green pixels from anywhere, but that is probably due to centroid sampling and other things which there are no settings for.
To clear things up, every DXT format outputs alpha, even DXT1. Like vtastek said, many of vurt's bark textures have random transparent pixels scattered into all the mipmaps, or even pine branch cutout alphas from another texture over bark. Since the models don't have a nif alpha property the pixels render as black normally. MGE XE doesn't have the benefit of seeing nif properties, so bark can become transparent. It still doesn't explain the white pixels, which is probably my fault. It could be a mismatch between the alpha test levels in the mesh and the fixed level distant land uses. Vurt's trees could also do with a contribution from a technical artist to clean up things that affect performance, because they are incredibly detailed but contain quite a few unnecessary flags and incorrect normals, making them a good target for optimizing.
It would be helpful if people tested toggling distant land, toggling shadows, and toggle mw/mge blending using macros, stating the mesh id and which options if any negate the sparkling. I'm going to work on an update for flora_tree_03 and flora_tree_wg_07 to see if mesh changes can cure the problem.
Ok so as i was testing the new dynamic lighting system i realized that all the AA flickering was gone! Can anyone else verify this as well? Also i have a few screenshots of graphical anomalies: http://s1095.photobucket.com/albums/i465/OldFashionedGuy/?/action=view¤t=MGEScreenshot001.jpg http://s1095.photobucket.com/albums/i465/OldFashionedGuy/?action=view¤t=MGEScreenshot002.jpg
What the hellllllll.