glthread: reduce dereferences of the next batch