The only way I have to test this is by running the mlockall.sh stress test, and testing with counters to count calls and cycles spent in vm_fault_copy_entry() indicates that this reduces performance slightly.
Before:
debug.counters.fault_copy_cycles: 66530662
debug.counters.fault_copy_calls: 90
After:
debug.counters.fault_copy_cycles: 68282538
debug.counters.fault_copy_calls: 90