I definitely agree that money and the market is and will drive NVIDIA to do what is needed.

In this case, an 8800GT is needed because:

1) The market wants a mid-range card that can actually perform. The 8600GTS is a joke and ATI just released the 2900PRO and will soon release the 55nm RV670. Jumping into the next high-end while leaving the mid-range market (where most of the money is made, as well as where most OEM contracts will be had) would be disastrous, esp. if you hand it away to your archrival. Keep in mind that more and more DX10 games are coming out soon (and have been now) and so the market for DX10 compatible cards is growing.

and

2) The 8800GTS 320 and 640 are the same core as the GTX/Ultra, but are likely the lower binned ones (with the lower binned shaders/streams disabled). But, the original cost to make each core is the same as the GTX/Ultra. The G80 core is HUGE and had record-breaking # of transistors when it was released. I'm sure that sucker ain't cheap to make and it might be smarter to go to a die-shrunk revised core that's cheaper to produce while leaving the expensive G80 core to expensive cards, the GTX and Ultra, and keep high profits.

and

3) Nvidia has always released a die-shrunk mid-range card before using that size for the high-end. This past year, a combination of the fact that the 8600 series was still 80nm, same as the G80, and that ATI tanked for most of the year, Nvidia feels that it can afford to release a mid-range card it needs, while using it as a testbed for a die-shrink for the high end performance crown, which it can release afterwards because there is no pressure from ATI to do so.

I might be completely wrong but from a business perspective, it can definitely make sense