Shrinking the heap by uncommitting regions in the safepoint takes a noticeable amount of time, particularly for large shrinks.

An alternative would be to move the slow part, i.e. the actual uncommit, into a concurrent operation.