Well, new drivers gave me an additional ~30FPS on Linux, but it's still not as close to the Windows machine as I expected it to be. Of course, there is a fairly large difference between the cards, so maybe it's not such a surprising thing after all. But I still have wierd glitches on Linux, so now I get to go through the code with a fine toothed comb, and maybe perform some hefty external tests with VBOs in other applications, to see if it really is a driver glitch and not a code glitch. But the engine runs rock-steady, smooth, and 100% glitch-free on the two different Windows installs I've tested it on (including the dual-boot install on the Linux box), so I'm pretty sure the code is okay. (Of course, the number of times I've been wrong about things like this vastly outnumber the times I've been right.)