http://bugs.winehq.org/show_bug.cgi?id=15789
--- Comment #10 from Vincent Povirk madewokherd@gmail.com 2009-04-07 14:12:45 --- I looked into this a little bit a while back and neglected to post the information on the bug. That means I have lost some details.
Profiling Worms: Armageddon showed that most of the CPU time was spent doing 8-bit -> 24/32-bit image copying (palette mapping). I believe this result is the same as from before the regression.
I also found that the copying was in response to ddraw functions like BltFast and Flip. I believe that BltFast would in the past have been an 8-bit -> 8-bit copy and not required until the final Flip when we do palette mapping on the entire frame. This leads me to believe that some surfaces are now being created as 24/32-bit, when they should be 8-bit.