Damn, Linus really knows his stuff, getting to the heart of the matter right the way - measuring user mode locking time is futile, as the OS scheduler can preemptively put the locking thread to sleep for arbitrary time. The OP ended up measuring how quickly the scheduler woke up the locking thread, thus measuring the wrong thing.
Yes, because the benchmark was using spinlocks incorrectly.
There are many places where the Linux scheduler is more responsive, even in gaming, exactly because of what makes spinlocks inefficient. The Ryzen scheduler debacle comes to mind.
