State-of-the-art Modelica tools for modeling and simulating multi-physical systems have reached certain maturity among the building physics community. Hence, simulation is widely used for control, sizing and performance assessment of energy systems. However, serious efficiency issues arise for large-scale models. This article proposes a practical application of co-simulation methods on detailed district energy systems. The aim of this study is to assess performance and scalability of co-simulation through Functional Mock-up Interfaces on a detailed and multi-physical district model. In particular, we propose a comparative analysis between classical simulation and co-simulation methods and a scalability analysis on a growing number of buildings. The models have been implemented using Modelica language and the OpenIDEAS library. A decomposition approach is taken for modeling the entire system, while stochasticity in the inputs is taken into account. Results are presented for various integration scenarios, including a classical integrated simulation for reference and co-simulations involving different master-algorithms within Dymola and DACCOSIM 2017. Scenarios are compared in terms of speed-up and accuracy of principal physical quantities representing key performance indicators such as indoor temperature, current and voltage at building's connection. The analysis shows that co-simulation can run up to 90 times faster than the integrated simulation for 24 buildings, while ensuring acceptable accuracy.