[danans]

Nothing at all.

> They examined [polymorphisms from] 2,000 people and correlated their similarity within the roughly 1,400 pairs of friends within the group.

>the Framingham data is relatively homogeneous in terms of ethnicity, consisting largely of people of European descent.

The population they studied seems very close knit and ethnically homogeneous, and the study's inferences are conditioned on this sort of population.

Presumably the consistent background genetics of the population allowed them to control for larger genetic differences and detect the subtle differences correlated with friendships in the group.

I would be surprised if the observed effect held up in a more cosmopolitan population where other stronger factors, both genetic and sociocultural, have more of influence in the formation of friendships. And, as others have noted, the observed effect was real, but very small.