A friend of mine spent last Christmas debugging an issue in memcpy in glibc (on Intel 32-bit CPUs). Glibc is less well tested than I expected, and has ASM implementations of many functions for many CPUs, some of which are (obviously) less well supported than others.
IA32 is probably not getting all the focus from devs and users this days, still surprising whoever... Do you have a link to the issue, out of curiosity?