In this study, we propose a novel server load balancing scheme. Server load balancing is known to be a critical mechanism in network-based information service. Most of existing schemes don't take servers' loadings into account, which might fail to make the loadings of all servers evenly and drive the server system to work on the brink of being over-loaded and/or out of function. The proposed scheme aims at preventing the occurrence of malfunction and saving the power consumption of the server system under a low loading while providing a better performance in response time. All connection requests are firstly distributed into one server until a pre-specified portion of its maximum allowed serving load is achieved. The following requests are then served by another server in the same way. The experimental results demonstrate the feasibility of the proposed scheme.