As noted at https://github.com/richfitz/storr/pull/111#issuecomment-510237712, this could conserve memory without sacrificing speed, and it could improve load balancing among threads.