For enterprise-wide and cross-enterprise workflow (WF) applications, the load of the WF servers and the amount of communication in the subnets may become a bottleneck. This paper shows how a distributed WF control can be realized in a way that the load of the components at run time is minimized. For that purpose, the control of a WF instance may migrate from one WF server to another. The WF servers are assigned to the WF activities in a way that minimizes the communication load. The server assignments are determined at build time by analyzing the WF model with respect to the actor assignments. As these actor assignments may depend on preceding activities, static server assignments are not always reasonable. Hence, so-called variable server assignment expressions are introduced, which allow dynamic server assignment without expensive run time analyses.