What I don't understand is since it's the same service can't we just run a performance test, determine the ballpark of the needed resources and just have them capped like in k8s? why would some servers be more powerful than others in the first place?