Efficiency in this aspect boils down to algorithms and functions as well, which are written by humans, although I'd argue overall skill caps in the industry increase. You can already check most planned hardware for the next few years. If you say we wait 20-30 years, that's gonna make a difference. If you are willing to wait that long not a problem at all, that's true, although standard consumer OSs become greedier as well ( WINstealurdata and overpriMACed ).
It's a simulation-
game, not a simulation. And by historical simulation you mean how history specifically happened in our case, or a including possibilities that didn't happen, if it's the latter the game profits a lot from well-abstracted general-systems, like the military-system or the galactic market, marginal work-productivity etc. ( usually =/= always ).
Also your analogy is wrong, dwarves or sauron do not exist ( they do as a concept, they aren't agents though ) in the real world hence it fails to be a simulation.
Also abstractions are quite important, get an MAS-lib and try to make a good stock-market sim that has remotely explanatory power. You judge without trying, you criticize without suggesting meaningfully.
There are things that are sub-optimal, the devs are aware of it La and Wiz said several times they'd rather do X. They have constraints, so they have to maximize locally and I am relatively certain their preferences are strictly monotonic( ous? Am not native ) when it comes features, and reducing computational complexity.