Scheduling stochastic workloads is a difficult task. We analyze minimum average response time of computational grids composed of nodes with multiple processors when stochastic workloads are scheduled to the grids. We propose an algorithm to achieve minimum average response time of grids. We compare the minimum average response time of grids with the average response time of grids with load balancing scheduling in different cases. Specifically, we analyze the impact of differential processor speeds, the number of processors per node, and utilization rate of the grids on the difference between these two scheduling strategies. These analysis provide deeper understanding of average response time of grids, which will allow us to design more efficient algorithms for Grid workload scheduling.