Given twice the number of CASs, about twice as slow is what I would expect for the mutex variant when uncontended. I don't know enough rust to fix it myself, bu

Given twice the number of CASs, about twice as slow is what I would expect for the mutex variant when uncontended. I don't know enough rust to fix it myself, but could you try with a spin lock? As the benchmark is very dependent on contention, it would give very different results if the the threads are scheduled serially as opposed to running truly concurrently (for example using a spin lock would be awful if running on a single core). So again, you need to be very careful to understand what you are actually testing.