For a multicomputer system without shared memory, we are modeling and evaluating different load distribution strategies which produce low overhead and require little or no a priori knowledge about the service-time requests of the jobs to be distributed. Homogeneous as well as inhomogeneous systems are regarded. The results strongly suggest that, while it is vital to take into account the actual utilization of the individual nodes in the distribution decision, little or no profit can be taken from the a priori knowledge of the workload characteristics in general-purpose computer systems.