As you should. My point isn't that it's ssoooo great that AMD seems to have a really healthy margin built into the 58xx cards. My point is that if the GT300 products are a good chunk faster - so what? It only really matters if they're a good chunk faster and *just as affordable*. Or at the very least, more expensive in proportion to the performance boost. The size and cost of the 58xx cards seems to indicate that, over the coming months, they have a lot of room to really come down the cost curve if they want to.
This time, G300 is launching after the competition. It's not like Nvidia doesn't know the price and performance of their competitors parts. Last time, they dropped their prices on the 280 and 260 very very soon after the 4870 / 4850 were launched, and even provided refunds through their card vendors. My point is, Nvidia is not going to commit retail suicide. They will price the card based on it's relative performance to it's nearest competitors. That's today's market reality, I would argue that they have no other choice. Discussions about how AMD will inevitably have this amazing price advantage do not seem to take this pricing history into account. Regardless of die sizes, yields, etc. etc., Nvidia will launch a product that is viable in the market, they have never done otherwise (or at least never failed to adjust prices quickly to keep with the current market).
AMD knows this, and will probably be OK with being just a little ahead on the price / perf curve, they're not going to greatly drop prices just because they can to the point where they are cannibalizing their own mid-range market.
As for the prices coming down fast on the high end, from what I'm reading Juniper is going to be pretty fast, and will be the part that really goes for the jugular, perhaps in a way that Nvidia can't respond to fast enough if they stick with a single product launch this year - i.e. if they have no new mid range DX11 part. I'm going to assume that Juniper will totally blow away G92 variants, making its natural competition G200 variants or better, which could be even more expensive for Nvidia. I don't see ATI dropping their high end parts to $200 that quickly - there's just no business reason to do it, unless Nvidia is able to totally bitch slap them on price / perf, which I don't think is too likely.
If Nvidia needs to price their high end parts so that they are price/perf competitive with AMD (at least enough to sell through their parts production), even if it's a 450mm^2+ part, that's doable. Trying to take a big loss on the mainstream parts is what will kill their bottom line, because of the much higher volumes. Personally, I am content to take their subsidized / price reduced high end parts, I don't really care about the mid range ones that much. This is the space where price perf really matters. In the high end I argue that overall performance is what wins, as long as it is not priced outside the realm of reason.
Another point I would make is that despite the excitement over the consumer parts, Nvidia currently makes most of their discrete GPU money in the pro card space. This is how they have funded G200 price parity (and development) despite the die size disadvantage, and I don't see that changing anytime soon. This is why I think Nvidia has the greater appetite for maintaining market share through price cuts in the consumer high end, and arguably a greater willingness or ability to take a loss in the high end consumer GPU space in a price war.