Testing software product lines (SPLs) is a challenging issue due to the combinatorial explosion of valid product configurations. To address this, similarity-based prioritization techniques are used to select diverse and representative subsets of configurations, maximizing feature interaction coverage. Although search-based test case generation has been widely studied, few empirical studies have compared prioritization techniques in such contexts. This paper presents a controlled experiment that compares five prioritization techniques: Novelty Score, Enhanced Jaro-Winkler (combined with Local and Global Maximum Distance algorithms), Dice-Jaro-Winkler, and a baseline technique that is based on Jaccard distance. All techniques were implemented in the PLEDGE tool and applied to the same initial test case samples to neutralize generation variability. The evaluation was conducted across nine feature models of varying sizes and three predefined test suite sizes. Results show that the Novelty Score technique consistently achieves higher pairwise coverage than all other methods. Enhanced Jaro-Winkler combined with Global Maximum Distance yields competitive results but exhibits performance variability depending on the feature model and test suite size. In contrast, the Dice-Jaro-Winkler technique underperforms across most scenarios. This study provides empirical evidence supporting the integration of Novelty Score-based prioritization in SPL testing and proposes an experimental framework reusable for future comparative studies.