It is most definitely possible to achieve great results, including performance, with Unity. It requires creating the game from the ground up with code, though -- and that does not seem to be the audience that Unity markets towards the most.
That said, it's not easy though it is certainly doable; it basically means you're using Unity largely as a "rendering pipeline" which may or may not be overkill, depending on the use case/game in question.
I know this is an old(er) post now, but thank you very much for the link. We actually had not seen that site before, and it's a treasure trove of fantastic information -- some of which we've found out ourselves, much of it new to us altogether.
You're right -- there's very little written on the topic. Though Unity has a fantastic community, and a large swath of information & user help to be found, a lot of it is directed at the newer/simpler developer (or mobile specifically).
Much of our work, and findings, has been derived from running our own tests & benchmarks as well as reading research/white-papers. In general, we've kind of found that many [advanced] game dev related topics are not well publicized or are fairly tightly-held. Our best source of information for things like highly advanced pathfinding, or low-level & high-volume multi-agent modeling/simulation strategies, have come directly from published academic research which has also been a treasure-trove of information.
We've read more research the last year or two than we did through all of school and the first ~10-15 years of our careers combined! =)
That said, it's not easy though it is certainly doable; it basically means you're using Unity largely as a "rendering pipeline" which may or may not be overkill, depending on the use case/game in question.