I did some work with testing a cross platform in-house library for read-write locking.

We tested cmpxchg16b and found the performance was terrible with more than 4 cores.

Ended up using spin-locks similar to Linux kernel RCU.