Multi-resolution is a useful tool for managing the complexity of huge terrain and geological data sets. Since encoding large data sets may easily exceed main memory capabilities, data structures and algorithms capable of efficiently working in external memory are needed. In our work, we aim at developing an out-of-core multi-resolution model dimension-independent, that can be used for both terrains, represented by Triangulated Irregular Networks(TINs), and 3D data, such as geological data, represented by tetrahedral meshes. We have based our approach on a general multi-resolution model, that we have proposed in our previous work, which supports the extraction of variable-resolution representations. As first step, we have developed, in a prototype simulation system, a large number of clustering techniques for the modifications in a multi-resolution model. Here, we describe such techniques, and analyze and evaluate them experimentally. The result of this investigation has led us to select a specific clustering approach as the basis for an efficient out-of-core data structure.