Okay, it's a fair point that '(lock) xchg' can potentially be faster than 'lock cmpxchg'. I wouldn't say it's really an ideal spinlock, because under real loads there are other considerations. That's why the Linux kernel uses queued spinlocks, for example.

For various reasons I still suspect that some other factor is at play, but I also don't really know how to tease that out. Maybe there are some performance counters one could look at, but I'm just speculating.