On further investigation (= a few hours staring at the ceiling last night), it looks like there's another explanation for my particular bug... which is good, because on further investigation (= a few hours squinting at google results) it looks like this probably can't happen on x86/x86-64 (I think? maybe?).

It's still true though that you can't just throw in a 'volatile' and expect cross-thread synchronization to work -- that's not C's semantics, and it's not true on popular architectures like ARM.