Current Internet and large-scale experimentation applications need to satisfy short provisioning delay and low blocking demands. Both can be guaranteed by using immediate reservation (IR) and advance reservation (AR), respectively. However, the scheduling of both reservation types in the same network can especially degrade the performance of IR if no extra policies are applied. In order to enhance such class-based policies, we need to quantify the future performance of the system, thus requiring to model its behavior. In this paper, we propose the use of a two-fold probability transition Markov chain successfully applied in the past in offset-based reservation systems. Results show the good accuracy of the model to simulation results, even in an scenario with multiple traffic classes defined by different book-ahead times. Such a performance validates its applicability to a wide range of immediate and advance reservation systems.