Single thread performance is hampered by multi threaded code.
Even when running in a single thread, 10% of the dWorldQuickStep() cycles are spent on locking/unlocking.
There is no locking need if running single threaded, so the cycles are wasted.
This should probably be addressed, so that the multi threaded mechanism does not hurt single threaded path.
Probably false alarm, as I did not use the '--disable-threading-intf' flag.