Your criticism is flawed.  You're obviously not from an engineering background.  If costs scaled perfectly for R&D, AM et al would all be on KnC's level of technology right now.  I compare the 40nm costs to those of smaller architectures using a roadmap analysis.  This is imperative because (1) as the difficulty continues to increase, the only way to match the required exponential increases in network hashing power to energy costs as the BTC price stagnates is to use more powerful (density-wise) chips, and (2) the efficiency engineering comes after you've completed a prototype at a smaller architecture - the "fine-tuning" of your draft chip design, followed by the hardware design process and the efficiencies that can be gained therein.

Your critique neglects to acknowledge that mining is an arms race and will be for the foreseeable future.  If you're fighting a war, sure you can bring a million musketmen to the battlefield, but they'll be a pittance compared to your enemy's cruise missiles.

Ha, quoting you right there! I don't necessarily think they'll achieve 15%, but why not at least 5%? I don't think that's totally unrealistic, maybe even 10%. I'm just saying, who knows what AM is doing, they may very well be putting massive amounts of hash rate online as we speak!