Today, many supermarket refrigeration systems are operated by decentralized control systems that often lead to excessive starting and stopping of the compressors which drive the cooling cycle and, consequently, to a large wear of the process equipment. In our previous work, a hierarchical model-predictive control scheme was proposed for supermarket refrigeration systems that overcomes this drawback. In this scheme, simple low-level temperature controllers are employed, and the high-level optimization task is the optimal adjustment of the parameters of these controllers. While this approach yields a good control performance, it is computationally too expensive for larger systems. In this paper, a more efficient approach is presented that is based on an approximation of the system dynamics using simple models that are computed from system sensitivities around simulated reference trajectories. The application of the new approach to a large hybrid model proves the real-time capabilities of the new technique.