Several different algorithms have been proposed in recent years for the dynamic optimization of resource allocation in virtualized data centers. The proposed methods range from fast and simple heuristics to exact algorithms that yield optimal results but take much longer. This paper suggests an algorithm portfolio approach in which multiple algorithms coexist. Based on continual monitoring and analysis of the state of the data center, the optimization algorithm that is most suitable is chosen on the fly. This way, the balance between optimization quality and reaction time can be tuned adaptively. Empirical results show that this approach leads to improved overall results.