I guess it is more accurate to talk about latency, because throughput can always be improved by accessing multiple cells in parallel.