There may be some scheduling delay, indeed. Another possibility is the result of OS caching. It seems to take a longer time the first time you try it, and then varying, but smaller times, on the next runs. For example, on the above command, all the subsequent runs lasted between 0.080 and 0.120 second, only the first one was significantly longer. Then, there is of course the effect of other processes running on the platform, but in the case above, I was the only user on the box, and not running anything else significant at the same time.