This line of reasoning makes absolutely no sense.
You said, "the benchmarks we've seen...have all," but there's been one benchmark that popped up. One. I do not understand how people are contorting that single anecdotal data point into thinking that there can't be a ML build.
Moreover, we are not talking about official or even sanctioned tests. This is an "oops" event by a person at Apple who should have known better but screwed up by running Geekbench. Why would you think that just because one guy is running a dev build on his laptop that there aren't other configurations/builds/etc. in use?
10.8.5 is in beta testing right now. It wouldn't be entirely trivial to be sure it's adapted for new models, but it's also not a massive undertaking. And, keep in mind--that's exactly what Apple did last year with the release of the rMBP. And that was even during a time when supply was seriously constrained.
Holding back the models for an OS release is just, well, insane from a business perspective. This is a company that is struggling mightily to avoid year-over-year declines in revenue. While supply chain issues are involuntary and cannot be helped, there simply aren't many good reasons to voluntarily delay your flagship, sales-leading product.
There has been 2 benchmarks with some time between them.
The first one made news on ALL the rumor sites.
So if it truly where just a mistake. There would be no chance that it would happen 2 times. One for 13" and one for 15".