This is an interesting article that summarizes some of the recent concern with the decline in effect sizes in a wide variety of experiments in many fields.
http://www.newyorker.com/reporting/2010/12/13/101213fa_fact_lehrer?currentPage=all
It also provides perspective on the long history of publication bias and how it impacts results.
My own feeling is that a lot of this comes from the definition of publication. In many cases people are not toiling away on their experiments in isolation, they are blogging, tweeting, etc about their work. So it is possible that the publication phases is arriving early in the process because of pre-print archives, blogs and the like. Patents, tenure, new drugs, higher search ad PPCs are not the only rewards for conducting scientific experiments that yield interesting results. As Facebook/Twitter illustrate clearly, the social rewards for sharing your ideas and getting feedback are enormous for most people.
As someone who has conducted hundreds of massive experiments in the context of search engine product development, I am going to renew my efforts to be sure that I am not kidding myself and amplifying the noise.
This article is not reason for panic, just a good reminder that proving things is hard and we should always be careful.
