Smoothing traffic flows at the network edge to reduce their burstiness has been shown to have significant benefits for video-on-demand systems and deterministic services. In this thesis, we investigate the relative abilities of smoothing and buffering to improve a network's admissible region for end-to-end delay-bounded statistical services. In single multiplexer systems, we show that buffering outperforms smoothing for any end-to-end delay bound and loss probability. We find that this behavior is due not only to statistical buffer sharing, but also to heterogeneity of the traffic flows' time scales. In multi-node scenarios, key issues for buffering and smoothing are user QoS requirements, traffic characteristics, and route length. For example, we find that as the number of hops traversed increases, the advantages of buffering diminish due to node-to-node buffer partitioning; and while smoothing is asymptotically superior, we find that in practice, the "critical route length" required to realize a smoothing gain is so large that buffering results in larger admissible regions, even in many multi-node scenarios.