By George, I think I've found it!

So, yeah, the problem definitely appears to be my mainboard. Memtest results:



...And a more animated version:



(Facebook folk try here: http://www.youtube.com/watch?v=J31ut0Bhv8w)

That's after about three days of running memtest. If it had been a RAM fault, memtest would have counted the error and continued. If it had been a CPU fault, it likely wouldn't have rebooted when I hit escape. (After it rebooted, it launched straight back into memtest, and that's running currently.) And if it had just been the SATA controller portion of the chipset, the fault probably wouldn't have registered at all.

So it's time for a new mainboard. I'm looking for something with two 16-lane PCIe slots, a 2GBx4 arrangement of PC1066, and support for the Phenom 9650. And it ought to be able to stand running with less than a couple hours down-time per month. Everything else is pretty much secondary.