Isn't this what automated unit and performance testing is for? If a refactor (or any code change) causes slippage on any of those indicators it should be flagged or rejected.