It depends on the calculus about (time) budget and stability. Maybe I consider performance already acceptable, and don't have time budget to investigate beyond that. Maybe I look at "nsync", see its mutex (may) change the fairness semantics, and then decide not to adopt it because this may break my callers; or its enough that it may change the fairness semantics, and I don't have the budget to test nsync or a new implementation based on the nsync algorithm to determine if the semantics differ.