This paper examines methods of optimizing energy consumption in pervasive computing scenarios where human users interact with infrastructure services. Workflow management systems can help in distributing these processes. With battery lifetime being a critical point in mobile applications, it is essential that the distribution is optimal with respect to minimal energy consumption. There already exists a distribution algorithm for which optimality has been proven for workflows but that only can be applied if the energy costs are known. Otherwise, the algorithm must be applied to predicted costs. The accuracy of the prediction clearly influences the result's optimality. We introduce several prediction methods and examine the question to what extend the result improves if the distribution algorithm can be applied repeatedly during the workflow execution. To this end, we augment the algorithm to deal with partly finished workflows. Finally, we evaluate the results with respect to predictors and prediction intervals.