In this paper, we investigate virtual machine (VM) scheduling problem for the downlink joint transmission (JT) of coordinated multi-point (CoMP) systems in centralized radio access networks (C-RAN). The inter-dependent computing tasks of users are centralized and scheduled on heterogeneous VMs to be executed. In order to achieve efficient scheduling, a priority based VM scheduling model is designed. Then, a delay-constrained and resource-constrained optimization problem to minimize the maximum completion time among users is formulated. Three scheduling algorithms including genetic algorithm (GA), genetic simulated annealing algorithm (GSAA) and packing algorithm (PA) are developed to solve the problem. Simulation results show that GSAA outperforms GA and PA in guaranteeing joint synchronization of all cooperating entities and fairness of users.