When reengineering large systems, software developers would like to assess and compare the impact of multiple change scenarios without actually performing these changes. A change can be ef- fected by applying a tool to the source code, or by a manual refac- toring. In addition, tools run over a model are costly to redevelop. It raises an interesting challenge for tools implementors: how to support modification of large source code models to enable com- parison of multiple versions. One naive approach is to copy the entire model after each modification. However, such an approach is too expensive in memory and execution time. In this paper we ex- plore different implementations that source code metamodels sup- port multiple versions of a system. We propose a solution based on dynamic binding of entities between multiple versions, providing good access performance while minimizing memory consumption.