Comment by haiku2077

I worked at at company near the top of https://en.wikipedia.org/wiki/List_of_the_largest_software_c... for a while. It was extremely common that web services only used about 1/20th of a CPU core's timeshare. These were dynamic web services/APIs. (We did have to allocate more CPU than that in practice to improve I/O latency, but that was to let the CPU idle to be ready to quickly react to incoming network traffic.)

This was many years ago on hardware several times slower than the current generation of servers.