I was agreeing with you that real world metrics would be useful.
I was making a couple clarifications. You seemed to imply that real-world metrics can be a substitute for big O, but that's not the case, they do different comparisons.
Also, I was pointing out that the 10ms vs 9ms example isn't a set of numbers that would indicate one algorithm is better than the other, just that one implementation is better than the other.
"a process that would take 10ms under the following conditions [a,b,c] now takes 9ms?"