The theoretical analysis of a cross-layer mechanism for improving the quality of service of real-time applications in wireless networks is presented. The mechanism coordinates adaptations of the modulation order at the Physical layer and the media encoding mode at the Application layer, to improve packet loss rate, throughput and mean delay. With the use of Continuous Flow Modeling, the system is considered as a "fluid" queue with inflow and outflow rates representing its traffic generation and service rates, respectively. Each data source is modeled as a Markov chain, from the steady-state of which the optimal adaptation thresholds of the cross-layer mechanism are derived. Performance evaluation results show that the optimized operation of the mechanism attains a significant performance improvement compared to both the sub-optimal mechanism, and a legacy system.