ParallelGC young collections sometimes need to allocate in the oldgen, and sometimes those allocations need to expand the oldgen. Currently that expansion is serial (under the ExpandHeap_lock), but most of the work is relatively quick. However, the optional pretouch of the expanded space is potentially long-running, possibly being a significant serialization point, as multiple threads come to wait for oldgen expansion. It would be better if those otherwise waiting threads could help with the pretouch.