Loop-closure detection is a critical component in LiDAR-based SLAM, as it corrects accumulated drift and improves long-term localization accuracy. While considerable research has focused on designing effective descriptors, the role of clustering and indexing methods in retrieving loop-closure candidates has received little attention. Existing systems primarily rely on KD-Tree, with limited evaluation of alternatives like DBSCAN and Spectral Clustering. This paper addresses this gap through a controlled, head-to-head comparison of KD-Tree, DBSCAN, and Spectral Clustering for organizing and matching LiDAR-Iris descriptors. All methods are evaluated under identical conditions using the same dataset, hardware, and binary descriptor format. Experimental results on a 333-frame dataset show that DBSCAN achieves the fastest total processing time (6.61 s), while Spectral Clustering offers the lowest query time (0.0386 s). In contrast, KD-Tree incurs higher memory usage (+631%) and the slowest total time (13.88 s). These findings demonstrate that DBSCAN and Spectral Clustering are viable alternatives to KD-Tree for loop-closure systems, each offering distinct runtime and memory trade-offs and provides practical guidance for selecting clustering strategies in LiDAR-based SLAM.