Look at the difference between bitfury and KnC. Bitfury has better power efficiency, even though they're 55nm and KnC is 28nm. The difference is bitfury is a 'full custom' ASIC while KnC is a standard-cell design.
It's likely that HF and Cointerra are using full custom designs. Their performance is probably pretty similar, and really fast. 16-20x the speed of ActM's FPGA copy design. I doubt it's because they use 16-20 as much die space.
Is this mostly the overhead of standard cell vs. custom or to a greater part due to the different design approach (if there is much difference, I have no idea)? Also 16-20x more space taken by structured ASIC vs. custom? I assumed there has to be a significant difference, but I never imagined it would be that great.