Identification of a neighbourhood that is based on multi-clusters was successfully applied to recommender systems, increasing recommendation accuracy and eliminating divergence related to differences in clustering schemes generated by traditional methods. Multi-Clustering Collaborative Filtering algorithm was developed for this purpose that was described in the author's previous papers. However, the solutions involving many clusters face substantial challenges around memory consumption and scalability. Differently, they do not equally take advantage of all the partitionings. Selection of the clusters to forward to the recommender system's input, without deterioration in recommendation accuracy, can be used as a precaution to address these problems. The article describes a solution of a clustering schemes' selection based on internal indices evaluation, that can be applied for input data preparation in collaborative filtering recommender systems. The results reported in this paper confirmed its positive impact on the system's overall recommendation performance, which usually increases after the selection of schemes. Additionally, a smaller number of clustering schemes on an input of a recommender system improves its scalability including memory consumption. The obtained values were compared with baseline recommenders' outcomes.