Imo, who wins the power consumption race will win the ASIC race in the long term.
The difference between the Gh/$ and the Gh/W is the difference between short-term and long-term decisions (buying the device vs operating the device).
In the short term I'm interested in buying the device with the best Gh/$ so I can recoup the initial costs faster. But what happens after I get my money back in 6 months, 12 months, etc.? Do I still mine with it? That decision will be made on power consumption costs, in other words I will keep operating the device if electricity costs per Gh are below profits obtained from a Gh mined.
In the context of bitcoin, terms such as 'short-term' and 'long-term' do not have the same meanings as we might normally attribute to them. For bitcoin, just a few weeks or months can have significant shakeups: BTC exchange rates, new products or services being introduced, major exchanges or businesses being hacked or exposed as scams, new government regulations being introduced.
Planning anything in the bitcoin realm beyond half a dozen months, let alone a year, is rolling the dice.
How long did CPU mining last once it became mainstream, before it was obsoleted by GPU mining? How long did GPU mining last before it was obsoleted by FPGAs? And by 'obsoleted' I don't mean 'unprofitable'. After all, GPU mining is still profitable (but CPU mining is not). I simply mean that a device has been introduced that is significantly more efficient in terms of GH/W and in terms of GH/$. And how long did FPGA mining last now that ASICs are around the corner?
All of these technologies (CPU, GPU, FPGA) lasted roughly a year. I was a small-time GPU miner back in July of 2011. I ordered BFL FPGAs as soon as they were announced. My first pair of Singles was mining in March, merely 7 months ago. I don't expect them to be mining come January, however.
Not that they will no longer be profitable ... they will ... but consider this: with a difficulty of 3 million, an FPGA Single makes $3/day while using $0.15 in electricity. This is quite efficient. Let's say that by the end of January enough ASICs will have been installed to drive the difficulty up 5x, to 15 million. In that environment, my FPGA Single will only be making $0.30 per day (1/5th due to difficulty increase, another 1/2 due to block reward dropping to 25BTC/block), but still requiring $0.15 in electricity.
So my FPGA profit will drop from about $2.85/day down to $0.15/day. Still a profit, but it would be silly to continue mining with such a device. Would I want to live with the heat and noise and wasting of space for a mere $0.15/day? Nope. I'd rather sell it. Or, if I couldn't sell it, I'd throw it away ... even though it was technically 'profitable'. Profitability is not the only criterion for determining whether to continue mining or not; there are many other variables to consider.
Same with GPU mining. GPU mining today is still technically 'profitable'. But to earn more than a few dollars a day after paying for electricity would require a large amount of GPUs. If they were quiet, and cool, and reliable, and didn't take up space, I would be happy to run them in a cupboard somewhere earning spare change. But they are nothing of the sort. I would not tolerate putting up with a bunch of GPUs unless they were generating significant profit. In today's environment, unless electricity is free, mining with GPUs is something I have no interest in.
Come January, FPGAs will be in the same boat as GPUs are today. As to my previous example of a BFL Single earning $0.15/day after the block reward cut and a 5x increase in difficulty: the ROI at $0.15/day would be ridiculous.
And back to OT, I would not personally be too concerned if the bASIC consumed 1.5x, or 2x, or 3x the power of BFL's offering; I think the 'short-term' profitability of all 1st-gen ASIC devices will be high enough that electricity costs (whether $0.10/day or $0.50/day) will be insignificant relative to gross profit. If you are looking for an ROI of 6-9 months, you will very likely achieve it regardless of power consumption. Higher power consumption would have mostly a logistical effect: how many I can run off a single wall outlet.
Only when the 'next generation' of hardware comes out, again with significant GH/$ advantage over these 1st-gen ASICs, will power consumption play anything more than a minor role. And is history is any indication, that could happen in the latter half of next year.