I wonder if you would see the same halting behavior if you stopped X and operated the machine solely from the shell session. You'd have to write a quick script that did some quick calculation ad infiniteum (add 1 to the previous sum, for example) and let it run for a few minutes, long enough to see if it did the halt thing.

If it works fine without halting, that would make me think that the trouble has to do with a package somewhere, and not the kernel. Then again, I go barking up the wrong tree with alarming regularity.