Comment 26 for bug 1079440

Note that the related launchpad bug is beginning to get some attention based on finding that even kernel 3.8-rc5 doesn't help.

I have a refinement for my workaround. I supposed that the important thing was to keep Xorg and compiz on the same CPU. But that is not the case. The most important thing seems to be to run Xorg on CPU 1. Note that my computer is an Intel Atom N280 whose two cores are is not advertised to be identical. I'd be suspicious of hairy signal timing based on this table. (Xorg 1 means "taskset" Xorg to CPU 1 etc.):

|| || Xorg 1 || Xorg 2 || Xorg unpinned||
|| Compiz 1 || WORKS || FAILS || WORKS ||
|| Compiz 2 || WORKS || FAILS || FAILS ||
|| Compiz unpinned || WORKS || FAILS || FAILS ||

All the cases that "WORK" have run with dual monitors and heavy use for more than an hour; many hours for the Xorg 1 cases. The ones that "FAIL" have never run more than a half hour or so under the same conditions.

Hope this helps.