Jeremy Berg's data with respect to scientific output and its linkage to grant funding is extremely interesting, and it confirms what I and many others have suspected all along, i.e. too high amounts of funding for a laboratory group can provide diminishing returns. This has enormous implications for the funding of mega projects, for example, from organizations such as Genome Canada.

From the careful analysis of the funding and scientific publications from 2938 investigators supported by the US NIGMS from 2007 to mid-2010, the Division of Information Services in the NIH Office of Extramural Research determined that the median annual total direct cost was $220,000, the median number of grant-linked publications was six and the median journal average impact factor was 5.5. As this covered a 3.5 year period, it would appear that the typical costs of a scientific paper with a 5.5 impact factor is about $128,000. However, in the estimation of these numbers, funding from non-NIH sources was not factored in, so the actual costs are even more. Frankly, this is an incredibly high cost, and it raises serious concerns about the efficiency of scientific research as it is funded and conducted today.

Looking at Dr Berg's graphs, it seems at first glance that productivity per grant dollar is dramatically lower in labs that are highly funded. However, as one commenter noted, it is difficult to separate funds for equipment and technicians from funds for scientific personnel; the data presented are insufficient to resolve the important issue raised.
As another commenter pointed out, lasting scientific impact is very different from high-visibility journal publications. The only other study I’ve read (about physics departments) suggested funding level was uncorrelated with lasting scientific merit.
These admittedly very imperfect measures are at least a starting point for reconsideration of current grant practices.