Comment by jesboat

Different applications can have very different requirements. I've worked on systems which would kill for 5% latency and on systems that would gladly pay 5% for better memory usage.