m1/m2 performance is only great while you live within it's limits. It's all special optimization tricks in hardware to match up with certain common software operations. Great if your usage pattern matches.

It's almost like when hardware has special bits that make benchmarks scream because they know exactly what the benchmark software does, yet normal work is the same or worse than anyone else. It's not exactly that bad. M1/M2 does perform a lot of real work a lot better. But it's "a bit like that"

It's just that it only works if your usage matches what Apple targeted as what most people will ever need to do.