In this paper, we describe the implementation and performance of a Virtual Audience behaviour model for Virtual Reality (VR). The model is a VR adaptation of an existing desktop model based on crowd-sourced data. The system allows a user in VR to easily build and experience a wide variety of atmospheres with small or large groups of virtual agents. The paper describes the model's implementation and possible benefits for different applications, e.g. training, therapeutic or culture. Our first scalability benchmark results demonstrated the ability to simultaneously handle one hundred virtual agents without significantly affecting the recommended frame rate for VR applications. Lastly, we propose a set of improvements our training system can benefit from. This research is conducted in the context of a classroom simulation software for teachers' training. CCS CONCEPTS • Human-centered computing → Virtual reality; • Software and its engineering → Virtual worlds training simulations.