Thanks for putting this together. I'm confident that J/gh will decrease enough to mitigate this, but the possibility that all mining hardware producers fail at making a more efficient miner is highly improbable (especially because there are so many now) but not impossible!
I am curious about your statement. How much more room do you think there is in this number? At 28nm everyone is getting about 0.8 J/GH. Going from 55nm to 28nm only reduced the consumption from 0.9 J/GH to 0.8 J/GH. Lets say we can make 18nm chips. The best case I would see is 0.5 J/GH and it would probably be higher.
In other words I don't see any way to get another order of magnitude change in this number - only smaller incremental improvements at this point.
Running the numbers at 0.5 J/GH is not a whole lot different than 0.8 J/GH and 0.5 J/GH may be a total pipe dream.
I believe half a joule is not unattainable for this year. 16nm is most companies' current focus, and some have designs underway for 14 and 10 (these are still 1+ years out, but it is something to note).
Running some quick numbers though, difficulty will increase at the same rate if not even higher. Taking all of this into account I firmly believe we will be paying over $5000 for a coin in one year.