Bitfury's developer is an amazing ASIC developer.  He also had efficiency unmatched by anyone else on FPGA and really only lacked a commercial success because of the false promises of BFL (both on FPGA side and early announcement of ASICs in a few months).   I would also point out that the rigs are essentially unavoidable underclocking.  They don't want to underclock but the chips are running slower than spec and that is going to improve the efficiency.  I am sure if you underclock or undervolt 28nm devices you will get improved efficiency (at the expense of less hashpower per $).

Still at stock clocks I don't think anyone is going to massively (<0.4 J/GH) outperform.  All the 28nm builders are taking pre-orders.  It is a competition for pre-orders and funding.  If they felt they could with high confidence say they can deliver <0.5 J/GH at the wall they say so because it would boost sales. 

Process node is only part of the equation.  KNC (28nm) is only guaranteeing 2.5 J/GH which is 4x worse than Bitfury despite being on a smaller process and BFL (65nm) is on a process very close to Bitfury but needs 5 J/GH which is 8x worse.  We can only go by what they unreleased specs and they are likely hedging their numbers to avoid an embarrassing miss but if KNC simulations were showing them 0.5 J/GH they wouldn't be building boards capable of 320W max for a 100 GH chip and only advertising better than 2.5 J/GH.  Still if Bitfury design can be shrunk to 28nm with similar efficiency it could be the most efficient chip yet.