Clock performance is sometimes referred to as granularity. The reason for this is because the clock performance details the smallest increment in time that can be accurately represented on a timestamp, (not the same as offset and sync).
In short, this is achieved by sending consecutive clock calls to the clock in question and measuring the difference between the two timestamps.
The graph above shows the EMA and the Raw (Raw output is the dots, EMA is the line). This is important when considering the application of accurate, synchronised time. This represents the average time it would take for an application to call the system clock and receive a response (a timestamp).
Although Clock performance is not the most used metric it can be interesting to understand, for example, if the clock performance spikes or is unusually high this can affect clock sync and as such can be an indicator that your device has a CPU utilisation issue or an alternative performance issue (just run a virus scan and see what happens, or try the following command to see how CPU performance affects your clock performance
cpupower frequency-set -g performance (you should see a decrease in clock performance measurements - this is good)
cpupower frequency-set -g powersave (you should see an increase in clock performance measurements - this is not good)