This paper investigates the effective computational resource allocation for large-scale fully-separable problems under the framework of a cooperative co-evolutionary algorithm called MLSoft. According to different subgroup sizes of the problems, we allocate different numbers of iterations to the subproblems in all the cycles. For high-dimensional subproblems, more iterations are needed during the optimization process; while for low-dimensional subproblems, fewer iterations will be assigned. The experimental results reveal that the proposed resource allocation scheme is simple but effective, which can enhance the performance of MLSoft in solving large-scale fully-separable problems. In addition, we conduct a group of experiments to evaluate the results if a higher weight is assigned to more recent performance in MLSoft. The results show that introducing weight to the latest reward affects very little on the performance of MLSoft.