To test this problem I have written a minimal windows application. If I force an access violation in the WM_PAINT handler this exception never gets to the debug
I've noticed that, when you have Aero enabled (which is by default in Vista), resizing windows tends to create many, many page faults. These aren't normal virtual memory-needs-to-be-paged-in faults, either. I suspect (though it's just a theory), that Aero redirects the graphics output to a protected chunk of memory, and catches the faults in order to know which bits of the visible surface need to be recomposited on the desktop. Perhaps that is eating other access violations as well.