By the way, if 9% in 3DMark2003 is the only big performance change (for any video card, whether it's just one or the whole product line), then that pretty much is an indictment of the uselessness of benchmarks.
If ATI can optimize a card or driver JUST for one specific benchmark, and I don't mean by detecting the EXE and "cheating", then it just goes to show how unrealistic those synthetic benchmarks are.
Always benchmark your real games to find out if there's a real world difference. Granted, you can't do that in every game, but you can get a pretty good idea with a few of the usuals.