Again, efficiency does not matter.

Look at it this way:

Assume every block is worth $50,000.

The amount of power that can be consumed to produce that block is $50,000 minus some overhead number (real estate, cost of hardware, labor, profit, etc.)  The higher the price the more power can be consumed.  Power consumption is directly proportional to price regardless of efficiency.

The more efficiently the power is used the higher the hash rate and difficulty of the network can and will be.  Efficiency drives difficulty and hash rate, price drives power consumption.

Sure some miners are better situated (cheaper power, cheaper cooling, better hardware, etc.) so they are more profitable than other miners.  This extra profit allows them to afford to buy even more hardware - pushing out the less efficient miners who will be losing money.  As new hardware comes on line and old hardware is retired the power consumption remains proportional to price.

Also, I am looking at the overall averages for the entire network and looking at estimating the power consumption of the entire network.  Local variations do not matter when looking at overall averages.