I agree, but practically speaking they're used en masse all across the field and people did bother to make them good [enough]. I suspect the benchmarks where they come up equal are cases where things are limited by other factors (e.g. syscall overhead), though I don't want to make unfounded accusations :)