I find it baffling that ATI still can't find their groove in the mainstream segment. And it's even worse considering the process advantage.
It's very predicable as they stick with those damned ratios...the instant HD2900 came out, I knew they would not have exceptional mid-range parts. It was obvious. Some people were saying "oh well high end isn't that important, if ATI nails the mid-range they will be better off" and back then I was like well guess what, they're not going to nail the mid-range, how could they? You immediately know they're going to castrate the TMU/ROPs, to more than likely 8 or less, which are the real limiter on the card, it's only a question of how badly. When your high end part has a bad ratio, then your mid parts will have a bad ratio as well. This is two gens in a row for ATI.
Nvidia doesn't really have a excuse for their lackluster mid-range though. If Nvidia was really managed as well as people act, there's nothing holding them back from dominating that segment, yet they didn't with the disappointing 8600GT. Although I think, you can make a case for Nvidia the 320MB 8800GTS might be encroaching on mid-range anyway, at ~$280.
I mean, HD2900XT is barely above mid range anyway. There's no way to cut it down and get good perfomance. Unless they left all 16 TMU's and ROP's in (as well as a decent amount of shaders of course), and we know that they're not going to do that. The only way they could really get a kickass mid range part though, is 16 TMU/ROP. 12 would give them something competitive, but not blow the doors off. Just like last time around.
The ratios totally killed them last time, as they aimed way way to low initially (X1600XT), then it took them a long time to recover, up the ROP/TMU count to something reasonable, and get decent mid range parts out. Which by then it mostly too late.