Bluetooth is a major technology for short range wireless communication. Scatternets expand its use to larger networks. If communication patterns are known before scatternet formation, frequently communicating pairs could be connected directly or with fewer hops to enhance performance. Communication patterns are obtainable by observing traffic, but this approach is ineffective in a pervasive computing environment where interactions are mostly spontaneous. We propose to use social group membership as an estimate of communication patterns. The proposed scatternet formation scheme forms a scatternet per a social group and then connects scatternets through tunnels. In order to ensure that each communicating pair has a reasonable share of bandwidth, the proposed scheme considers the number of hops and branches and traffic distribution when interconnecting scatternets. Simulation results show that the proposed scheme not only enhances the total throughput of scatternets but also shares network capacity more fairly among communicating pairs.