Weighted Fair Queueing (WFQ) scheduling algorithm and its variants could be used to provide real-time guarantees by making bandwidth reservation. However, on the one hand, hard guarantees are based on peak workload model that leads to underutilize network resources. On the other hand, soft guarantees, based on average workload model, are not always sufficient to maintain acceptable Quality-of-Service (QoS) since consecutive packet losses or deadline misses may occur which is not suitable for real-time applications. We propose in this paper a trade-off between hard and soft real-time guarantees to maintain an acceptable QoS in overload condition and maximize efficiently the utilization of network resources. The key of our solution is based on the fact that many real-time applications, such as voice and video, are loss-tolerant. The loss profile must be well defined and such a profile can be easily specified using (m,k)-firm model. Therefore, we propose (m,k)-WFQ algorithm to take into account (m,k)-firm timing constraints to provide delay guarantees of at least m packets out of any k consecutive packets without violating bandwidth fairness or misusing network resources. Using Network Calculus formalism, analytic study gives the deterministic delay bound provided by the (m,k)-WFQ algorithm for upper bounded arrival curve traffic. We extend our analytic results for guaranteed-rate networks, such as IntServ QoS model and ATM networks, and DiffServ QoS model. Theoretical results and simulations show a noticeable improvement on delay guarantee made by (m,k)-WFQ compared to WFQ without much degrading bandwidth fairness.