The future Internet will support a large variety of services whose QoS requirements may vary widely. In order to provide scalable quality of service traffic aggregation may be necessary. We examine several approaches to providing scalable QoS services which includes soft QoS with relative performance guarantees, and hard performance guarantees signaled by RSVP over an aggregated cell switched core. In these approaches QoS is based on aggregate traffic. The question then arises as to what level of QoS is perceived by individual flows within the aggregate. This has an important implication on QoS guarantees which the network can make to individual users. To provide answers to the above question we analyse per-flow and aggregate QoS in an ATM virtual path (VP) which is used as an example of a cut-through handling IP flows. Using cell loss as a measure of QoS, our analysis allows us to investigate the impact of the parameters of individual flows on per-flow QoS. We then compare per-flow QoS against the QoS experienced by the aggregate flow in the VP. Based on numerical tests we find that cell loss experienced by individual flows may be very different from cell loss experienced by the aggregate flow. Thus if aggregate cell loss is used as a QoS constraint in dimensioning of VP this may lead to unfair treatment of flows and hence to potential user dissatisfaction.