The idea behind virtualized audio is to extract sound sources (performers) from their native acoustic spaces and insert them into a virtual space, in which there are one or more listeners. This paper focuses on the insertion step, which is an interactive application that exhibits high computational demands from both aperiodic and periodic real-time tasks, has minimal communication demands, and exposes a significant number of adaptation mechanisms. These properties make it well suited as a driver application for research in interactive high performance computing on shared distributed environments such as clusters and grids. We describe the application, our implementation, and the adaptation mechanisms in detail, and then present an initial performance evaluation of one of its components, and study different server selection strategies for it.