We study the problem of multiple-packet bundling to improve spectral efficiency in cellular networks. The packet size of real-time data, such as voice over Internet Protocol (VoIP), is often very small. However, the common use of time-division multiplexing (TDM) limits the number of VoIP users supported because a packet has to wait until it receives a time slot, and if only one small VoIP packet is placed in a time slot, capacity is wasted. Packet bundling can alleviate such a problem by sharing a time slot among multiple users. A recent revision of cdma2000 1xEV-DO has introduced the concept of the multiuser packet (MUP) in the downlink to overcome limitations on the number of time slots. However, the efficacy of packet bundling is not well understood, particularly in the presence of time-varying channels. We propose a novel quality-of-service (QoS) and channel-aware packet bundling algorithm that takes advantage of adaptive modulation and coding. We show that optimal algorithms are nondeterministic polynomial time (NP)-complete, recommend heuristic approaches, and use analytical performance modeling to show the gains in capacity that can be achieved from our packet bundling algorithms. We show that channel utilization can be significantly increased by slightly delaying some real-time packets within their QoS requirements while bundling those packets with like channel conditions. We validate this paper through extensive OPNET simulations with a complete evolution-data optimized (EV-DO) implementation.