While stereoscopic 3D visualization is reported to have positive impacts on the delivery of visuospatial information, little is known about its cognitive mechanism. This study investigates the cognitive load incurred during mental rotation tasks with eye-tracking, comparing the stereoscopic 3D stimuli with 2D equivalent both presented by Microsoft HoloLens2. The study also explores the separability of different cognitive aspects using multiple fixation metrics: total fixation time, average fixation duration, and fixation frequency. While all three parameters differentiated easy/difficult questions, the difference between correctly/incorrectly answered questions was only seen in total fixation time and dispersion-based fixation frequency, and the difference between 2D/3D questions was only observed in average fixation duration and velocity-based fixation frequency. Results suggest that 3D representations could be perceived as more detailed helping users create better mental images for visuospatial operations, and that these observed eye-tracking parameters reflect different cognitive aspects.