Let me put it another way: GT200 will not be economically viable within its lifespan.
We are not talking about GT200, GT200b will be more than viable. See most of the things people are saying now of GT200 were said about G80 and R600 in the past, especially of the latter. Then came G92 and RV670 which had almost the same architecture, but had big improvements in the fab process (many things in the marchitecture related to the process too) and were more "optimized": narrower memory controler, improved ALUs, etc.
Very few has been said about GT200b (confirmed by Nv, I couldn't care less of the rumors). We only know it will be 55nm. A die-shrunk GT200 they call. I remember a time when G92 and RV670 were also called just a die-shrink. And was true, to an extent, but they were a lot more. This "news" tell us that GT200b might have GDDR5 and DX10.1, and that would mean it will be a lot more than just a die-shrink. Much more than what G92 was to G80.
As to why they didn't make G92 in the first place and why they have repeated the "mistake" with GT200, I do have my theory. If you want my opinion, G80 was as it was and GT200 is as it is (same concept, more extreme), not because they were looking for gaming performance alone, but also for other aplication's performance too: CUDA and (IMO to more extent than what most may think) TESLA. I'd bet TESLA is as important for Nvidia as the workstation market is for Intel and AMD. Enthusiast here (and elsewhere) have the tendency of overestimate desktop market and underestimate the bussiness market. They will base their next TESLA on GT200 as they did with G80 (no G92 TESLA) and it requires some things that graphics don't require.
Those things (common claim: "You don't need a 512 bit interface, it doesn't affect gaming performance") are what made G80, R600 (was also made with GPGPU in mind) and GT200 very big and expensive, a better balanced for gaming chip is coming soon to fix that just as G92 came in the past. Until then you have what you have, and you can buy it or not. They don't have to please you all the time, they sell their product so you can buy it, but they don't owe you anything. Sincerely, people need to understand Nvidia and Ati (and Intel, AMD, etc...) are companies doing their bussiness, they don't owe us anything. In the case of Nvidia, GT200 is the product, which they made to implement on GTX cards as well as in Quadro and TESLA solutions, GT200b will only be desktop and fit better that (and only that) role. But GT200 on it's own is a good product, better than the competition in many ways except the price, if you don't like it don't buy it, but in no way it is comparable to the FX series. In fact, contrary to FX, GTX cards are faster than the competition, while being better in perf/watt, heat output and overclocking. Even only by die-shrinking it (without the aforementioned optimization that IMO is inevitable and was on Nv's mind from the start) they could fix the perf/price, because it will probably allow both lower prices and better clocks.
In the end all that I said is speculation, as we don't know anything and I don't necesarily believe in it. I just wrote it to counter your speculation. That way we stay neutral, your message is it will never be viable and my message is it could kinda own again. The real thing will be somewhere in the middle.
You're comparing the highest model (ATI) and the second highest model (NVIDIA) to eachothers. If we look at the difference between HD4850 and GTX260 you'll see that it's 3%. That fits well inside the error margin, so I wouldn't call that "quite better". The difference between HD4870 and GTX260 is 7%, not exactly groundshattering IMO. There might be a little advantage in perf/watt for NVIDIA, but it isn't big enough to turn the tables for g200 (especially consedering the size of the die -> perf/price).
Unlike you, I'm comparing the right cards. You have to compare the cards on the same performance level, no matter from which brand they are or where is their place in the lineup. Comparing GTX260 to HD4850 is like comparing and sports car to an utilitary. And 280/4870 like comparing the Ferrari Modena to a Maranello. Of course they will consume more they are faster and it's a lot more expensive (in $ and consumption) to increase performance the higher you go. Same with GPUs. There are physical limits and constraints on the perf/watt matter and because of that, the higher you aim the worse it will be. Nvidia aimed higher AND using a bigger fab process, it would be natural if they had a lot worse perf/watt, but in reality they have the better one. 55nm wil only increase that advantage. BTW that advantage seems to be architectural as it was present in G92/RV670 too. Look at the charts, G92 owns.
(it was a big jump in technology)