Probably card dependent, and seems to at least in part be mining software dependent.

 It DOES look like the best efficiency point on EBWF is at 104 watts not 106, but it's so close it could be an artifact of measurement limits.

 I tried pushing core and memory harder on one of my EVGA SC cards, but stability went bad very quickly above the +200 core and +700 memory settings (which was an unreported part of my original testing).
 That however might vary with the specific card and probably with the model of GPU.


 I'm not waiting, as I'm not convinced that the Volta versions of the 10xx series cards will be a LOT more efficient.
 10% probably, 20% maybe, 30% I'd bet against - we're still on the same process node, just "process improvements" and perhaps some additional minor optimisations to the designs are all we're likely to see - think of "RX 4xx vs RX 5xx" level changes being the most likely.

 Hint - "12nm" isn't a new node, it's enhancements to a current 16nm node process with a "new name" as strictly an ADVERTISING point.


 Samsung does seem to have some VERY small quantity of production happening on 10nm - but strictly for low-power devices like their smartphones at this point, and they're eating ALL of the output so far.
 Intel is supposed to FINALLY be shipping their first 10nm products next year - but the date keeps slipping.....

 Manufacturing efficiency for both is probably going to be VERY poor on anything they manage to ship at 10nm next year - it takes TIME to dial a new process in, and it seems to take longer every time a legitimate "shrink" happens TO get the new process dialed in.