Read kernel code to see how CPU utilisation is calculated. In essence, count scheduled threads to execute and divide by number of cores. Any latency (eg. wait for memory) is still calculated as busy core.