third run, standard deviation, sanity check and even quieter mode
kcbench compiles the kernel twice with various number of jobs and prints the results; the user then has to check which of the settings was the fastest. That's suboptimal.
Improve this by (1) detecting which number of jobs seems to be the optimal one for the particular system. Then (2) calculate the standard deviation and (3) yell if anything looks fishy (like one run being a lot slower or faster than the others).
Once done a (4) even more quiet mode is possible which only prints the the best result, as long as it's not that much faster than the other two runs which used the same number of jobs.