A general class of optimal fault-tolerant partial-connection multiple-bus networks and their self-routing algorithms has been developed. By conducting a large number of network simulations, using self-similar traffic we compare the performance of these networks with those of traditional single-bus networks and complete-connection multiple-bus networks. Our simulation results show that given a fixed number of processors, these partial-connection networks have improved performance over single-bus systems, perform comparably to the more costly complete-connection multiple-bus systems, and have flexibility in setting the performance level and fault-tolerance level. Moreover, we demonstrate that, in certain of these networks, if the system load increases resulting in decreased performance, this can be handled by adding buses to the system rather than completely changing the topology. Thus, for very little additional cost and no change of topology or routing algorithm, increased loads are handled.