Designing island models is a challenging task for researchers. A number of decisions is required regarding the structure of the islands, how they are connected, how many individuals are migrated, which ones and how often. The impact of these choices is yet to be fully understood, specially since it may change between different problems and contexts. Cluster geometry optimization is a widely known and complex problem that provides a set of hard instances to assess and test optimization algorithms. The analysis presented in this paper reveals how design options for island models impact search effectiveness and population diversity, when seeking for the global optima of short-ranged Morse clusters. These outcomes support the definition of a robust and scalable island-based framework for cluster geometry optimization problems.