Any time I read an article that references the 41 shades of blue experiment in a negative context, it turns me right off the rest of the article.
Testing 41 shades only sounds bad because we typically have far too many design variables relative to the population of users. In those cases, we need someone with a priori knowledge of good design to limit the space we explore. When you can make billions of observations, you can start exploring experimentally much more of the design space.