Given the problems with the 40nm process its not suprising they ditched the efforts on the 200 refresh and concentrated on the 300 series... its gonna be a pretty painful time for them but once they crack it - it should pay off - in the long run its the better design as you get better performance returns as you scale upwards compared to ATI's design which is going to start seeing exponential diminishing returns in the long run without a redesign.
EDIT: Having said that look at what happened with the P4 and Athlon... the P4 is kinda like the ATI cards and the Athlon like nvidias design (in a way)... the P4 ran out of headroom but the stronger Athlon design for whatever reason didn't carry them into the next generation very well whereas intel came back with the core 2... so it could go either way.
You really can't get that any more backwards, Nvidia is a pure and utter speed demon, theres no efficiency, its brute force, it uses 50% more transistors, almost double the clock speed in shaders, yet comes out, about the same performance. AMD are getting similar performance from half the clocks and half the transistor space.
Sorry but considering the P4 was bigger with a shedload of cache and massive transistor count relying purely on clock speed to compete with a complete lack of efficiency, even having a double pumped fpu(or was it an interger part, I can't remember that far back) which is even closer to Nvidia's core clock speed essentially double pumped shader clock speed.
Nvidia HAVE ALREADY hit their ceiling, they've already had issues increasing clocks, they've got so many transistors at such a high speed they can't produce the thing, yet AMD push forwards, with on time releases of significantly faster cards compared to the previous generation, with very few large problems, with an incredibly efficient design that will continue to improve in efficiency. Remember Nvidia is at peak efficiency, its got zero headroom, you take a 285gtx and optimise the drivers, theres no headroom it can largely use its entire 240 shaders at any given moment. AMD has the ability to increase performance pretty much exponentially as game dev's program in a different way.
Its almost unbelievable you could be that naive as to quite literally get it completely and unquestionably backwards. Remember the 280 and the 285gtx both missed their target clocks, just with less of a problem than Fermi has had, but a far smaller clock increase on the 285gtx than planned as their architecture which most certainly does not lend itself to smaller processes gets worse the lower they go.
Fermi simply can not and will not succeed in the future. If they have a deritive of Fermi at 28nm, with the same design and try the doubling all the features route, it will simply not release. AMD would have zero issue releasing a double numbers on everything part on the next process node, why, because they've been watching manufacturing and design a architecture based on the problems being faced, while Nvidia have ignored what almost literally every single other gpu/cpu/chip maker in the industry is doing, which is finding leakage a major problem. Everyone else in the entire industry has moved away from raw clock speed and brute force, to efficiency and paralelism, every single last one but Nvidia.