Distributed publish/subscribe systems are poised with challenges of performance degradation and poor scalability. This is typically caused by an uneven load distribution of real-world applications and the susceptibility of link failure in networks. Partitioning and replication techniques have been implemented by exploring community-based load balancing to cope with such issues. The novel approach herein exploits offloading at the inter-community level as well as filter replication at the intra-community level. This results in the dynamic distribution and forwarding of publication and subscription services among brokers during run time. The proposed method, Co-Lab (COmmunity-based LoAd Balancing), seeks to improve the network performance by clustering brokers in a community by taking into consideration interest similarity and filter replication. It attempts to effectively achieve a more consistent and uniform load distribution among brokers and to circumvent the occurrence of highly overloaded brokers. Performance evaluations indicate thatCo-Lab has promising advantages by achieving relatively better load balance, reduced overall load and robustness against failures.