This paper investigates the minimisation of the computational cost of providing Immersive Voice Communication (IVC) for mobile clients attached to a Distributed Virtual Environment (DVE). When experiencing IVC, each listening avatar in the DVE, receives a mix of the surrounding avatars’ voices, all rendered according to their respective positions in the virtual world. We propose to deliver our mobile IVC service using a type of server-based network architecture to perform voice spatialisation function and mixing for each mobile client. This model will minimise both the access bandwidth and processing power requirements of the mobile devices. We propose a model for computation reduction in servers that re-uses the result of audio spatialisation for several clients by allowing certain level of voice localisation error. Using Linear Programming (LP), we have developed two optimisation formulations to examine different scenarios of applying the computation reduction scheme, with different avatar densities and under varying rendering accuracy constraints. The insights gained from our analytical results should benefit future IVC service providers for mobile DVE.