Methods and apparatus, including computer program products, for workload balancing in environments with multiple clusters of servers. A process for sharing resources from a pool of servers among a first cluster of servers and a second cluster of servers includes determining a current workload of the first cluster and determining a current workload of the second cluster. The process calculates a combined workload of the first and second clusters, and allocates a first number of servers from the pool to the first cluster, the first number equal to a product of a total number of servers in the pool and the first server's workload divided by the combined workload.