https://bugs.freedesktop.org/show_bug.cgi?id=94133
--- Comment #4 from John Bridgman john.bridgman@amd.com --- I think the answer is "not knowingly". Optimizations tend to be done for an entire generation (or a few generations), not for specific chips.
That said, if a specific chip is observed to be an outlier (performance quite a bit lower or higher than expected) then that specific discrepancy will probably get some attention.
I don't understand your Mafia2 numbers - the Linux version wouldn't be running under Nine, would it ? If the Linux version were running under OpenGL, then this would look like a fairly typical "DX renderer in the app gets a lot more work than the GL renderer" scenario.