Scalability is a main obstacle for applying Model-Driven Engineering to reverse engineering, or to any other activity manipulating large models. Existing solutions to persist and query large models are currently ine cient and strongly linked to memory availability. In this paper, we propose a memory unload strategy for Neo4EMF, a persistence layer built on top of the Eclipse Modeling Framework and based on a Neo4j database backend. Our solution allows us to partially unload a model during the execution of a query by using a periodical dirty saving mechanism and transparent reloading. Our experiments show that this approach enables to query large models in a restricted amount of memory with an acceptable performance.