The parallel iteration support supplied by OopStorage scales poorly to multiple threads, particularly when the work to be done for each block is small, as may occur when the storage contains weak references.  For example, a microbenchmark for JNI weak handles such as from JDK-8176454, with JNI weak handle processing parallelized, shows little or no parallelization speedup.

This seems to be a consequence of a combination of (1) contention when attempting to claim blocks, and (2) each thread touching the majority of blocks as it walks the block list.