From thew point of view of a capacity planner and performance engineer, one goes for low latency first, because for many algorythms, the best throughput is seen when latency is lowest. Some cases will be different, but starting with removing time wasted sitting in a queue is A Universal Good (;-))