Internet worms pose a serious and ongoing threat to system security, often resulting in significant service downtime and disruption. In recent years peer-to-peer (p2p) networks have become a target for the deployment of worms as their high connectivity allows for rapid dissemination and homogeneity of the adopted software platform ensures the existence of common susceptibilities. In this paper we observe that peer similarity in p2p networks can greatly increase overall vulnerability; peers with largely different system characteristics are unlikely to be infected by the same worm. With this in mind we present di-jest - an autonomic method for neighbour selection based on heterogeneity. Our results show the efficacy of di-jest in reducing the spread rate and potency of p2p worms. By selecting neighbours with different system characteristics di-jest can reduce the number of peers infected by a worm by up to 80%.