Taking 5830 out of the equation for a moment, GF104 *should* yield better than Cypress all things equal. Even if it isn't, there are even more things to consider.
Not only is it slightly smaller but nVIDIA has made some interesting choices:
While a 5850 has 90% of it's shader modules and all of it's ROPs, the gtx460 has 87.5% of it's shader modules, with the option to use those with or without a faulty ROP block. Granularity aside, this already puts them at an advantage because of options.
Also, nVIDIA is using low clocks. Partly because of GF100, but it also allows salvage of more chips. While we see it as overclocking room, for nVIDIA it means chips can widely vary and still be used. 675mhz @ ~1v is not asking a lot, especially since nVIDIA chips are not binned at a specific voltage like ATi's chips are.
For reverence; That is the the same clock and ~voltage as CEDAR, ATi's smallest part, that one would assume is clocked for maximum yields; certainly not performance.
Finally, nVIDIA always could put out a 6SM part. While 5830 has 70% of it's SMs/50% of it's ROPs, and such an nVIDIA part would be a square 75%, ATi also uses a high voltage so the chips are viable at their needed speed of 800mhz; needed bc so many ROPs are disabled. Because nvidia is binning to 24 ROPs, a 288sp part is still viable at low clocks, unlike 5830.
Last I heard (which admittedly was a while back) Cypress yields were somewhere around 70%. I'm -ASSUMING- that was for 5850/5870. I imagine things have gotten better, but one must also assume things will be as good, if not better for nVIDIA because of the aforementioned binning 'advantages'. I'm not doubting the author's 80% number -guess or inside baseball- as it sounds feasible, but nVIDIA's should be high as well.
In short, nVIDIA's plan of setting the bar low for GF104 to allow for greater yields, causing in-turn lower stock performance and market value, made up through volume is actually a REALLY good idea; especially considering the fact they still need to make GF100 look like a viable part until it's off store shelves. It is practically the anti-thesis to ATi's plan, and while different, it's GREAT for consumers.
ATi - smallest chip possible to control yield/price, highest reasonable stock clocks for performance.
nVIDIA - widest-market chip as possible for performance, low-clocks to control yield/price.
Coming from an ATi fan, you still have to recognize GF104 is a damn good chip. A full 8SM SKU could for all intents and purposes replace GTX480 if clocked like an ATi part. It would also probably cost $400 like an ATi chip. Instead, for now, nVIDIA will likely go the route of clocking it lower to steal 5870's thunder...and I ain't gonna complain irregardless of their motive.
For now, I think nVIDIA deserves a little credit...
...We can all laugh later when they rebrand it with a higher clock/2GB to compete with Southern Islands.