Heuristics are often used to support model composition, but they also lead to syntactic and semantic inconsistencies in the composed models. If the effort to resolve inconsistencies is high, heuristic model composition might become impractical. Unfortunately, little is known whether well-designed models can minimize the inconsistency rate so that state-of-practice heuristics can be efficiently applied. This paper presents an exploratory study that analyzes the impact of model stability on the composition effort required to produce several releases of a software product line. Our results, supported by statistical tests, show that when models are well-structured upfront and, therefore more stable over time, the inconsistency rate and inconsistency resolution effort are kept under control. On the other hand, when changes are not predicted upfront, the use of existing heuristics might become prohibitive.