Cloud Computing applications are inherently oriented to send and receive information between multiple virtual machines and geographically distributed users. That exchange of information produces a delay with a big impact on the overall application performance and in the quality of service perceived by the users. This article presents a Tabu-Search based algorithm to distribute the virtual machines in multiple data centers with the objective to increase the quality of service, and also considering the energy cost. A network with more than 1,000 nodes is simulated, and 50,000 applications are added using more than 12,000,000 virtual machines. The results report an average delay that is 35% of the delay obtained with the First Fit policy, and a reduction of 18% in the energy cost.