State-of-the-art Modelica tools are very effective at converting declarative models based on differentialalgebraic equations into ordinary differential equations. However, when confronted with large-scale models of distributed systems with a high number of states (1000 or more) or with large algebraic systems of equations (1000 or more unknowns), they face a number of serious efficiency issues, that hamper their practical use for system design. The paper analyses these issues in detail, points out strategies for improvement, and also introduces a library of scalable test models that can be used to assess existing tools, as well as to help developing advanced solution methods for large-scale systems.