The complexity of an (ontology-driven) conceptual model highly correlates with the complexity of the domain and software for which it is designed. With that in mind, an algorithm for producing ontology-driven conceptual model abstractions was previously proposed. In this paper, we empirically evaluate the quality of the abstractions produced by it. First, we have implemented and tested the last version of the algorithm over a FAIR catalog of models represented in the ontology-driven conceptual modeling language OntoUML. Second, we performed three user studies to evaluate the usefulness of the resulting abstractions as perceived by modelers. This paper reports on the findings of these experiments and reflects on how they can be exploited to improve the existing algorithm.