mining profitability – Would AsicBoost have made “all the difference between a profitable and an unprofitable mine” “for the bitcoin mines of the future”?

Back when AsicBoost was a thing – the paper by Timo Hanke, “AsicBoost – A Speedup for Bitcoin Mining”, made the following claims in the introduction:

Through gate count reduction on the
silicon ​_AsicBoost_ (claim 1:) ​improves two essential Bitcoin mining
cost metrics simultaneously and by a similar factor: the energy consumption (Joule per Gh) and
the system cost ($ per Gh/s). With the system cost being proportional to the capital expenses of
a Bitcoin mine and the energy consumption being proportional to its operating expenses, (claim 2:)
AsicBoost​ reduces the total cost per bitcoin mined by approximately 20%. (claim 3:) For the Bitcoin mines
of the future​ AsicBoost ​will make all the difference between a profitable and an unprofitable
mine.

Claims 1 and 2 are correct (ignoring, for claim 2, the specific 20% improvement specified, just that any improvement would reduce the total cost per bitcoin mined).

But claim 3 is untrue due to the fact that the difficulty would change to adjust to the faster hash rate.

  • Correct?
  • Correct only as long as AsicBoost was “secret” and in use by only a small percentage of miners?
  • Correct only insofar as it was “secret” to a miner that was thus more easily to assemble as majority of the hashrate and attempt a 51% attack to change the rules?

(I guess, as a secondary matter, and having just learned about it now, post SegWit, I’m wondering what the fuss over AsicBoost was. And what I have gleaned is that it is because it was kept secret and could have been secret long enough to allow someone to (more) easily attempt at 51% attack.)