The test setting is important here - the test is on an otherwise idle machine. This means that the update ensures that some thread wakes on a timer every second which may explain the large drop. This test is interesting, but not very representative of a real world usage scenario. It’ll be interesting to compare it to the results of the other test they running, where they keep a video running in the background.
I'm still a little curious of what's causing the increase in power use. A single additional wakeup per second should not have a two-digit percentage impact on power use when even an idle machine is probably going to have dozens of wakeups per second anyway. I wonder if updating the seconds display somehow causes lots of extra wakeups instead.
