It's not always about productivity. It can be about finding superstars. These two ideas contrast: in productive organizations, you typically want people closer to the mean (i.e. reduce variance among the population) which yields constant improvement. In an organization that yields superstars you want people far from the mean (i.e. a lot of variance among the population) which yields big hits but not necessarily constant improvements.

Basically, the organization (scientific community) needs to decide which type of population to select for to fulfill their strategy.