i don't buy that at all. 100x means that the person could do one day's worth of work that would then take five months for a "normal" engineer to accomplish the same work. that simply isn't realistic. there may be one off cases, but that's a possibility in any field, although you seem to claim otherwise.
and "orders of magnitude" implies a lot. 10x is one order of magnitude, and 100x (highly doubtful anyway) is just two orders of magnitude.
Playing football, what is so hard about it, you just run and kick the ball right? Then why some players are orders of magnitude better than the others?
Even stocking shelves can be optimized if you would like, I bet I could stock shelves after some optimization 10x or 100x faster than average workers. (as a student I was working stacking shelves)
If you or anyone can stock shelves 100x faster, and the skill wasn’t copiable easily, they would for sure earn a lot more than the average worker. Hell, they can patent it and earn royalties.
Proper mental models, tool mastery, love for hyperfocus, brutality of testing, passion for solving interesting, difficult problems, along with the proper work environment involving a proper problem.
The same could be said for Andrew Wiles' process of solving Fermat's Last Theorem, tho in a vastly different arena.
If I had to nail it down to a single characteristic, it would be obsessive tenacity in the pursuit of a high standard of craftsmanship. From that, all else comes with time.
Unlike in other types of say, physical labor where a top performer is just a few (X) more productive than the average performer -- in software, a top performer can be orders of magnitude (10-100X) more productive / more creative than the average worker.
I don't know what exactly contributes to this (intangible knowledge turning into action much more concretely, ability to avoid important pitfalls much more capably, ability to demonstrate correctly much quicker, no dependency on physical action, etc), but I have observed this to be a phenomenon.