Early designs of mobile ad-hoc networks (MANETs) assumed they could be scaled to very large topologically flat networks where every node in the network has equal capability and access to the wireless channel. This assumption was called into question in 2000 by Gupta and Kumar [1] who established an asymptotic theoretical limit for node capacity as the number of nodes in a single channel MANET becomes large. Recent field experiments have confirmed that this theoretical limit may be difficult to overcome. To address MANET scalability alternative MANET designs have been developed in which a wireless channel are shared by only a subset of the entire collection of network nodes. Subnets are then linked by gateway nodes that operate on two or more channels and provide connectivity across the entire network. In these MANET designs gateway nodes transmit and receive on multiple subnets and can become bottlenecks in the network if there are too few gateways. On the other hand, if the network contains many gateways, then too much subnet bandwidth may be allocated to gateways and insufficient bandwidth allocated to single channel nodes. This paper examines the design of MANETs based on two subnets and determines the optimal number of gateway nodes which maximizes overall network performance for a uniform message traffic loading.