This paper presents a performance evaluation study of the scalable reservation-based QoS architecture. This architecture introduces a scalable per-flow signalling model, using techniques and algorithms developed to minimise the computational complexity, namely a label switching mechanism and an efficient timer implementation. The underlying architecture is based on DiffServ and resource reservation is performed for aggregates of flows at both core and access networks. The obtained results show that this architecture is able to provide QoS guarantees, irrespectively of the behavior of other flows in the same and in different classes, maximizing the network resource utilization. Based on the performance evaluation, we can state that this architecture is able to support service classes with strict and soft QoS guarantees in high-speed networks.