The Fork/Join task ForEachOps.ForEachOrderedTask can be ...:

1) Instantiate the ConcurrentHashMap with a size of Math.max(16, AbstractTask.LEAF_TARGET << 1), to avoid resizing that can result in slower concurrent operation on the map

2) Evaluate whether the the reporting of elements needs to be within a synchronized block to guarantee a happens-before relationship, if so synchronization should be applied consistently and perhaps the code should be cleaned up as follows:

           if (task.getPendingCount() > 0) {
               Node.Builder<T> nb = task.helper.makeNodeBuilder(
                 task.helper.exactOutputSizeIfKnown(rightSplit),
                 size -> (T[]) new Object[size]);
               task.node = task.helper.wrapAndCopyInto(nb, rightSplit).build();
               task.spliterator = null;
           }
           task.tryComplete();
       }

       @Override
       public void onCompletion(CountedCompleter<?> caller) {
           if (node != null) {
               Node<T> _node = node;
               node = null;
               // Dump any data from this leaf into the sink
               synchronized (lock) {
                   _node.forEach(action);
               }
           }
           else if (spliterator != null) {
               Spliterator<S> _spliterator = spliterator;
               spliterator = null;
               // Dump any data output from this leaf's pipeline into the sink
               synchronized (lock) {
                   helper.wrapAndCopyInto(action, _spliterator);
               }
           }
           ForEachOrderedTask<S, T> victim = completionMap.remove(this);
           if (victim != null)
               victim.tryComplete();
       }