As usual, you are very well informed, Guy!
To be fair, with 0.16 J/GH this would be the most efficient chip in the wild so far.
The 0.6V BM1384 value is just quadratic extrapolations of the 0.66V value. Try it, it fits perfectly. IMHO too perfect for a measured value.
Anyway, 21 really screwed it up with the GH/mm2 value. In best case they have here about 2 GH/mm2 (for sure not with 0.16 J/GH). This is a good GH/mm2 value for 28nm, but for 22nm it should be 3 ... 4 GH/mm2. Otherwise this chip will be simply too expensive in production to be competitive for standard miners.
You have got to stop thinking about conventional economics here. The IoT bandwagon is impervious to conventional economic considerations. After all, folks are going to just be so happy to pay an extra $25 for their light bulb, it doesn't matter if the ASIC price is a little high. The fact that it's going to be part of the IoT relieves it from those pesky metrics.
If it's connected to the Internet, it MUST be great!!!
Isn't Raspberry PI 2 + Sidehack's 1384 dongle is essentially the same for quarter of the price ?