Last night, while trying to compile the ReiserFS quota patches into the 2.4.17 kernel, I got segfaults. It has happened before on lesser hardware, but this time, I suspect I should have paid a little more to get the ECC Memory. From The SIG11 problem:
However when the hardware occasionally flips a bit, gcc uses so many pointers, that it is likely to end up accessing something outside of its addressing range.