Games like EverQuest and Dark Age of Camelot occasionally produce natural experiments in social science: situations that, through no intent of the designer, offer controlled variations on a phenomenon of theoretical interest. This paper examines two examples, both of which involve the theory of coordination games: 1) the location of markets inside EverQuest, and 2) the selection of battlefields inside Dark Age of Camelot. Coordination game theory is quite important to a number of literatures in political science, economics, sociology, and anthropology, but has had very few direct empirical tests because that would require experimental participation by large numbers of people. The paper argues that games, unlike any other social science research technology, provide for both sufficient participation numbers and careful control of experimental conditions. Games are so well-suited to the latter that, in the two cases we examine, the natural experiments that happened were, in fact, perfectly controlled on every relevant factor, without any intention of the designer. This suggests that large games should be thought of as, in effect, social science research tools on the scale of the supercolliders used by physicists: expensive, but extremely fruitful.