This reminds me of when my boss set me to look at why certain counts varied so much. I thought it was electrical variation, but he thought it was heat. Turned out he was right. I then looked at how we were measuring the counts. They were distributed over a narrow range in a bell curve. Someone had set the gate so one side was on the top of the curve. Heat shifted the distribution slightly and this setup was sensitive to slight shifts. When I widened the gate to include all of the curve (to 3 sigma), the variation disappeared. When people are looking at effects that are close to the 2 or 3 sigma limit, all sorts of crazy but meaningless variations can occur. Then people publish papers on cold fusion, etc.