In Web services environments, classes of service are offered at different performance levels. Requests belonging to various service classes are to be handled differently in order to yield their expected corresponding performance level. We investigate a mechanism that provides this service discrimination by deploying an optimized external scheduler, as opposed to common mechanisms for prioritization and internal server scheduling. A measure of Service Performance Level (SPL) that combines target and achieved performance values is introduced. Our approach is based on simple Weighted Round-Robin (WRR) scheduling with dynamically adjustable weights that are computed by solving an integer resource allocation problem. We present experimental results, contrasting a few scheduling policies. Further, we provide an analytic queueing network model to approximate the behavior of the system, and show its effectiveness.