A spread-spectrum channel can accommodate severa concurrent successful transmissions, and a single-transceiver node can thus utilize only a small fraction of the channel's capacity. As a result, the maximum network throughput is much lower than this capacity whenever a single node, such as a gateway or a file server, must carry a large fraction of the traffic. In order to allocate the appropriate fraction of capacity to a "busy" node, we propose to equip it with several transmitters and receivers, thereby turning it into a "supernode". Several architectures and operation policies for supernodes are suggested and compared. It is shown, for example, that an M-receiver supernode can significantly outperform M independent conventional nodes. In a slotted system with packet lengths of one slot, this is achieved by special routing of the supernode's inbound traffic. In an unslotted system, it is achieved by appropriate code assignment policies. Packet-ratio networks with half-duplex nodes, as well as networks with full-duplex nodes, are considered. © 1988.