In a data center, networks can have lower latency than disk (even ssd). Now the real place this all falls on its head is page faults. That there are definitely places where you need to have granular control over what can and cannot stall a thread from making progress.