> There is absolutely no doubt that all those sites, and many others, are continuously running such experiments.
They are running randomized experiments, but they are not necessarily running the right experiments. If the bottom line is revenue, then attrition may not be directly measured; if they are testing on per page-view or per session instead of bucketing users permanently, then they can miss effects too; and if they are testing by bucketing users simultaneously, they can miss non-additive site-wide effects. If you look at my comment, I link two publicly-documented randomized experiments, and both set up/analyze their tests in an unusual way.
After I published my results, I heard about some parallel results, but not that many. I heard whispers that Google might (or might not) have internally run equivalent tests; I heard from Chris Stucchio that an unspecified company he worked for had run an A/B test with similar results; I heard from a mobile developer that they had run a similar experiment but found no effect of ads; Pandora subsequently published their A/B test; and Mozilla has published this. I also read through most of the 'advertising avoidance' literature in academia going back 30 years, and found nothing relevant. And that's it.
So, if everyone is always continuously running obviously relevant experiments... where are they?