WiredTiger recently got an external contribution from Github user AGSaidi to use the -moutline-atomics flag for ARM in the cases where the compiler supports it.
In GCC-10, this flag is on by default. The functionality has apparently been backported to older versions of GCC however it needs to be explicitly set with the -moutline-atomics flag. On my local machine, I saw wtperf ops/sec increase by 2x on a few of the workloads that I tested.
If you're already aware of this flag, feel free to close this. I figured I should let someone from the server org know just in case.