Scheduling schemes have important effects on the performance of load sharing in distributed systems. Because the load sharing policy based on CPUmemory has made the system memory an important role in effecting the system performance, it can reduce the paging fault and enhance the usage of the system resource. In light of the characteristics of the load sharing based on the CPU-memory and the variation of the jobs in executing, new CPU local scheduling schemes named more memory-request more CPU Slice based on Round Robin (RR-MMMCS) mechanism and more memory-request more CPU slice based on predicting (MMMCS-P) mechanism, are presented. Furthermore, the effects on the load sharing policy based on the CPU-memory of the variance of the interarrival time and service time are discussed. The tracedriven simulations show that the load sharing policy based on the CPU-memory and RR-MMMCS, MMMCS-P scheduling schemes are effective and have better performance in average response time for both CPU-memory and memory-bound jobs.