This paper aims at maximizing the profit associated with running geographically dispersed green data centers, which offer multiple classes of service. To this end, we formulate an optimization framework which relies on the accuracy of the G/D/1 queue in characterizing the workload distribution, and taps on the merits of the workload's decomposition into green and brown workloads served by green and brown energy resources, respectively. Moreover, we take into account not only the service level agreements between the data centers and clients but also different deregulated electricity markets of data centers located at different regions. We prove the convexity of our optimization problem, and the performance of the proposed workload distribution strategy is evaluated via extensive simulations.