Accelerating retain/release is important, and a really big clue as to why M1 is so fast, but does nothing to reduce the amount of memory a program needs.
Gruber acknowledges that, as everyone else does, in a vague seemingly tautological way that amounts to "if it really needs more RAM, it really needs more RAM."
But his claim isn't about that, it's about equal performance with less RAM, or faster performance with the same RAM. In that regard, speed does matter.
Most programs can run with varying amounts of RAM. If machine A has 32GB, it might run program X with 16GB, say. Machine B, with only 16GB RAM, can still run X, but might use only use 10GB. If all else were equal, A would prrobably run X faster. But if B has speed advantages in RAM use and perhaps other ways, it might run X just as fast or faster.
That's the main point, as I understand it. Just to clarify what I think Gruber means about a couple other points:
He mentions that Android uses garbage collection and iOS uses reference counting, and that's one big reason why an iPhone needs less RAM than an Android phone. This is true! Of the two memory management techniques, GC tends to be more wasteful of RAM. But it's a non sequitur here. We're only concerned with macOS on Intel versus macOS on M1, and the frameworks aren't going to count references any harder on M1 than they do on Intel. Apple has given us no reason to believe that memory management at this level is any different on M1.
His point there is only to illustrate the principle, not to claim the same particular difference applies with Intel, though that might confuse some people who aren't reading as carefully as you are.
Gruber mentions that he has a lot of stuff open on his 16GB review unit, and everything's smooth. But that's not an impossible result for Intel Macs either.
He may be comparing to machines comparable in cost, size, etc. The comparisons I've seen are remarkable, though not yet rigorous as you point out in your other post.
