I haven't pursued this as a conscious policy but I can confirm that every time I've done it I've always discovered something that would have invalidated the benchmark if I wasn't looking at it while it ran. Article mentions it but I've seen something as simple as "gee that's strange that should be using all my cores not just one" several times over, or equally stupid things.