Nothing will give publishers a better idea of how to configure things than actual experiments. I would suggest running each such experiment for at least a week so that the data is statistically relevant. Also, if an experiment is to be useful, it must not contain more than one type of change compared to the previous setup, so that the difference in earnings can be properly attributed. Depending on the nature of the experiment, publishers may also want to leave each experiment running for longer in order to make sure Google has had enough time to re-index the album pages. This may not matter if the change only involves ad colors or placement, but it will most certainly do if the keywords targeted by the text and metadata associated with the photograph have been affected. Remember that in serving contextual ads, AdSense relies on an offline analysis of a page, rather than performing that analysis every time ads have to be served on that page. Looking at the ads served for a page that has been changed but for which AdSense has not yet updated its cache may be quite confusing.