Existing experimental research on behavior in weakest-link games shows overwhelmingly the inability of people to coordinate on the efficient equilibrium, especially in larger groups. We hypothesize that people are able to coordinate on efficient outcomes, provided they have sufficient freedom to choose their interaction neighborhood. We conduct experiments with medium sized and large groups and show that neighborhood choice indeed leads to coordination on the fully efficient equilibrium, irrespective if group size. This leads to substantial welfare effects. Achieved welfare is between 40 and 60 percent higher in games with neighborhood choice than without neighborhood choice. We identify exclusion as the simple but very effective mechanism underlying this result. In early rounds, high performers exclude low performers who in consequence 'learn' to become high performers.