[avar]

Isn't this always going to be less efficient than profile guided optimization?

Of course that assumes you can profile your program with realistic "production" workloads, and it'll need two compilation passes, so having sensible ML defaults boxes sense.

But it's odd not to compare this to PGO in terms of the resulting performance.

[jeffbee]

It's not odd because the result is on top of PGO. You would never adopt this approach if you did not already have PGO. The gains from PGO will be much larger.