We consider a set of n heterogeneous servers which differ in processing time and Quality of Service (QoS). Jobs are divided into m levels with regard to their service ranks. We present a Threshold Policy (TP) depending on number of different jobs in the queues as a practical and flexible dynamic routing policy to control the QoS. Two performance measures are discussed: the QoS and the Average Waiting Time (AWT) in the queues. The TP is compared with both a static routing policy which maximizes the QoS level and the Minimum Expected Delay (MED) policy which minimizes the AWT. Numerical example validate that the proposed TP is more effective when both measures are considered. The TP balances the trade-off between QoS and AWT and therefore it is superior to the MED policy and any static routing which keeps the QoS in a certain level.