Consider the large update:
against an empty triple store, when the actual update is very quick and makes no changes; and the vast bulk of the time goes in the optimizer.
In two separate shell windows execute:
One of these, for me, executes in 35 seconds, the other in 70.
This indicates that there is a synchronization somewhere in the server. Since, from ticket 893, we see that virtually all the time is going in the optimizer, and this work need not be exclusive, something is a little awry