And it would be good for the mining operators too, because they'd pay less for electricity!
That's now how mining works.
You can look back at the advance in hashing power per watt and you will see that despite constant progress the actually energy consumption has still grown.
A more efficient s21, for example, will simply make people buy 11 of those instead of 10 as for the same amount of power you can get a bigger slice from the reward.
Bitmain could launch a miner that makes 1Petahash for 2kW, the consumption will not go down once they roll out all those miners, it's just that the hashrate will explode and that's all.
What can make the consumption drop is a price decrease, and although I pray it doesn't happen we might test it by the end of the period, below 40k any s9 generation miner running with more than 10cents/kwh is useless, the s11 will follow up at under $35k.
Bitmain and co. can only manufacture a limited supply of miners and I expect that their manufacturing throughput of hypothetical energy-efficient miners would remain the same or close to it.
So it doesn't really matter if one mining ops buys more of those miners than the old ones, because they're buying at the expense of other buyers. Global usage of ASICs stays about the same, assuming that the manufacturers keep selling out as they frequently do.
Since only the mining farms know how much of the time they keep their ASICs on or how long they will use them for, we currently can't really estimate their total energy usage for sure other than for the inefficient (and probably nonsensical) worst case of leaving them always on, and it's still unlikely this'll be possible if what I call "green" ASICs are ever deployed, but the energy decline should be noticeable in say 10 years when most of the old miners are phased out.