How it comes the Watt/GHash of the final product is not much better with 65nm process. Doesnt it suggests the future 23nm ASIC miners would not be much more power effecient thus today ASIC will still compete fine with future ASIC?
Good question. BFL were bragging that their technology was so much more advanced than "the competition", then when they finally got working silicon (after several expensive design iterations!!), they discovered that it was far more power hungry than their estimations, and had to redesign their boards (and whole product range, note that they are currently only shipping their smallest Jalapinos). Just too much heat being generated inside their cute little boxes.
As for why? Well a like-for-like comparison for the same hashing architecture on the 65nm vs 110nm process should give a significant efficiency gain (I've been out of this biz for a long time now, so I won't hazard an exact figure). So it looks to me that Avalon did a better design (its Ngzhang after all, who did the Icarus and Lancelot FPGAs, he's good).
As for 23nm, that will be a while off (if ever) as the mask costs are far more expensive. Expect next gen to be 65nm (possibly with improved architecture), then maybe 45nm if
BTC is still with us by then.
PS I forgot kncminer. They are using a 28nm process but its all still vaporware at the moment. They have takena very few pre-orders for September delivery (they are using a mask sharing scheme to save on up-front costs, so their initial throughput is very limited). We'll just have to see what comes of it.