Given that the 5770 uses about 108 Watt and the 6990 375 Watt (around 3.5 factor), yet also performs at around that fraction/factoring, it appears that the performance per Watt actually scales.
That is incredibly impressive to me as a feat for AMD/ATI.
The initial purchase cost still is at a premium (100 vs 520 EURish, so more a factor 5 than a factor 3.5), but compared to not requiring a whole new "case" around it, I guess you have a fair argument there.
(Although you could probably build a barebone supporting 3 5770s for 2-300 EUR, too)
Cool, learning something about the GPUs on the side here. :p
Sorry for going off topic so far.
Btw - I am not sure you need the 16x PCIE..given the nature of this stuff, it isn't exactly that you are constantly swapping gigabyte of texture data per second..but then again, I have no idea what the bandwidth throughput is on P-miner(though I suspect it is negligible).
Would be worth looking into though, seeing as how it would enable 3-4 cards vs 1 per PC.
(I.e. maybe just keep one of the 5770s for a normal PCIE and run it parallel, see how it performs? If still 180 mhash/s, then there at least my theory holds)