Various algorithms have been proposed for the problem of quality-driven service composition. They differ by the quality of the resulting executable processes and by their processing costs. In this paper, we study the problem of service composition from an economical point of view and adopt the perspective of a Composition as a Service provider. Our goal is to minimize composition costs while delivering executable workflows of a specified average quality. We propose to dynamically select different composition algorithms for different workflow templates based upon template structure and workflow priority. For evaluating our selection algorithm, we consider two classic approaches to quality-driven composition, genetic algorithms and integer linear programming with different parameter settings. An extensive experimental evaluation shows significant gains in efficiency when dynamically selecting between different composition algorithms instead of using only one algorithm.