For the DX9 games I played, it was a noticeable jump to go from the 5900xt to a 9700 non-pro, which are cards that were mostly on a par in their day.
I played a lot of Battlefield:Vietnam when I switched from the 5900xt to the 9700. When I took FRAPS data from my normal play and set all framerates above 85 (monitor refresh) equal to 85, I found the 9700 non-pro could get the same framerates that the 5900xt could at 1024x768 at 1024x768 4xAA 8xAF or at 1280x960 0xAA 8xAF.
Capping the framerate was important, because the 5900xt would do great on simple stuff and hit like 120+ FPS when the 9700 may still be under 100, but it would really suck in places, dropping the framerate into the TEENS. It made for a horrible experience unless you turned options way down. Capping the framerates more closely tied my benchmarking to my actual gaming experience. It was the very low framerates I was noticing (negatively) and was the reason I went searching for another card.
Here is an example for HL2:
http://www.xbitlabs.com/articles/video/display/half-life_15.html
Look at 1024x768 4xAA 8xAF, the 6600GT is like 30% ahead of the 5900, and this is with AA and AF, where the extra bandwidth of the 5900 should be giving it an advantage... AND the 5900 is running in DX8.1 mode because it sucks at DX9.
I guess what I'm saying is the 5900 series simply doesn't cut it for DX9 games, and you should notice a significant difference between it and a 6600GT. I am a firm believer in Average FPS don't tell you everything.