While GEostationary Orbit (GEO) satellite systems provide us with a wide coverage area, their long delay serves as a significant constraint for real-time applications. On the other hand, Low Earth Orbit (LEO) satellite systems are best suited to delay sensitive applications. However, the coverage and mobility issues of LEO satellites lead to relatively high management costs. In this paper, we devise a new load balancing and quality of service (QoS) provisioning scheme to accommodate both real-time and non-real-time traffic based on a new congestion-prediction scheme. The effect of this new scheme is expected to improve the efficiency of the GEO/LEO hybrid satellite networks and the QoS satisfaction of end users.