Computation and communication are the primary dichotomy of looselycoupled multiprocessor resources. Modeling and simulation are the techniques used to estimate machine performance based on the speed of these resources. The accuracy of the resulting performance estimates is often questionable, since such techniques cannot usually take into account all of the system detail: They become intractable in effort. Any real, hardware implementation of a multicomputer immediately fixes the speed of its resources, and can only yield a single point on the performance curve. We have implemented a third technique, called time dilation[2], to evaluate the performance of loosely-coupled multiprocessors by varying the ratio of computation speed to communication speed. This technique requires a high-speed clock and a test multicomputer system. Time dilation provides a way to measure accurately the performance of a given program on a variation of the physical transport system of a real machine.