http://bugs.winehq.org/show_bug.cgi?id=22000
--- Comment #5 from Austin English austinenglish@gmail.com 2010-03-12 11:09:57 --- (In reply to comment #4)
Partially fixed. The bad indices were an internal Mesa issue and I've already committed a fix, but the segfaults remain. There's definite heap corruption during the d3d9 tests, during context destruction. Steam straight-up segfaults while trying to launch Civ4. I think the two are related and have the same origin, but I haven't nailed it down yet.
Have you tried running with 'WINEDEBUG=+heap make device.ok' ?