Diminishing returns of science due to complexity barrier

I like the recent Arxiv article by Claudius Gros, where he shows that the positive results of applied science for the society (like increase of life expectancy, or accuracy of weather prediction) are sublinear (and thus convex) functions of the invested resources, due to the increasing complexity that arises in an already advanced research field or technology. Particularly nice is his simple mathematical conclusion that funding agencies could therefore make a more efficient use of the total amount of money by distributing it to many smaller science projects, rather than using up most of it for a few mega-projects (such as the LHC).