Yes, that is correct. If disk access will be done in the (write) lock
then the lock is released until after the disk access. In 2.0.x this
happened in a good number of places and in 2.2.x it will happen almost
everywhere (we have a new exception system to generalize this
behavior).
This leads to very high concurrency even with just the greedy writer
(reader/writer) lock since it is just cpu/memory bound.
More granular locks will be good, but these yields on long running,
slow or disk operations already provide good performance.