It is interesting to see the math behind why it did not works and probably it is a good approach to never discard an idea just because it sounds stupid. But I see really a little value in this line of thinking.

One of the big issues with experiments is how much value they really drive, which is the real effects. In many situations we do A/B testing for changes whose real impact is a fraction of a percent of the tracked metric. If anyone followed the discussion about p-hacking in science [1] should really start wondering if the way we run experiments, considering the little changes we see, is not nearing p-hacking, especially when I see this ideas floating, and if we don't need to find more significant way of running our experiments.

[1] One of the first article I found about the issue http://journals.plos.org/plosbiology/article?id=10.1371/jour...