Large groups often face problems that require both the discovery of a solution and its subsequent coordinated implementation. Since information sharing in groups usually occurs through interpersonal interactions, properties of the group's communication network may impact collective performance. We present results from a behavioral experiment consisting of 49 trials conducted over 4 sessions that examined the connection between performance and two changeable characteristics of a group's network: the diversity within the subunits that comprise the group, and the centralization of its communication topology. In our task, a large-group adaptation of a classic small-group network experiment conducted at MIT in the 1950s, groups of 20 participants attempted to solve a problem that required them to pool unique information, and then disseminate the solution to others. We found that networks comprised of subgroups containing diverse information were more efficient at both discovery and dissemination of the solution, even in cases where homogeneous subgroups could theoretically complete the task in an equal number of steps. Contrary to the original small group results, we found no evidence of a positive association between collective performance and the extent to which connections in the network are concentrated in a few central individuals. We conclude by discussing the implications for organizations and the extrapolation of findings from small to large networked groups.