3 ms·

We're not talking about the general case. One thread per core might be optimal for that specific workload.