You didn't mention how you're measuring this.
Pieter's dont seem to show that, and estimates of a finer timescale than that are going to be very noisy.
Anything that reports the network's hashrate is a very very vague estimate as it can only be observed indirectly by block times, and blocks are somewhat infrequent. Some reporting tools miscalculate the hashrate around retargeting because they apply the wrong difficulty (the earlier block's difficulty) or because they switch to a single block integration interval.
There is very little a miner can do to influence the hashrate by turning on and off for just a brief period... and it doesn't really matter where in the cycle they do it. Moreover, whatever influence they achieve is just balanced out by the next cycle. There should be no way to increase your income from doing this, and it's very easy to decrease it (by letting other people get more of the blocks).