Satisfying Quality of Service (QoS) of multimedia traffic has been gaining more attentions recently. Data, voice and video share the same communication channel but have significantly different traffic characteristics and performance requirements in terms of packet loss, delay, etc. Self-similarity also imposes some new challenges for network dimensioning as more buffer and bandwidth are expected to accommodate the extra burstness of self-similarity of multimedia application. However, previous studies did not investigate how much extra bandwidth is needed to meet QoS requirements of self-similar multimedia traffic. In this study, we model delay and loss requirements of interactive voice as the constraints to simulate the maximum allowable link loads. Results of this study suggest that under a self-similar input traffic, most of the bandwidth resources of slow links will be wasted to accommodate the traffic burstness if quality requirements of voice traffic are to be satisfied in a best-effort manner. For fast links (>10Mbps), self-similarity seems to have little impact on the maximum allowable load, which is above 82%.