So it took only about 3.3X MH/s/W to drive GPUs out of the business of SHA256d mining altogether. Is that right?
Check your math, please. Antminer S9 is 13 TH/s using 1300W or 10000Mh/s/watt or 3300X of your GPU number.
Th/s is 10^6 Mh/s, not 10^3 Mh/s.
Therefore, your assumption is off by a factor of, say, 1000 (10^3)
In other words, a single S9 is working (in BTC ) as a 3300 GPUs.
Since Z9 is only 10-12 times more efficient in Zcash than GPU, you second assumption is not correct either.
You did slow down the ASIC advantage from 3300X to a “meager” 10-12X (so far). IMHO, it is high time to make it 1X.