While haptic interactions are increasingly integrated into embodied Extended-Reality (XR) educational systems, their impacts on students' learning outcomes remain underexplored. To address this gap, a comprehensive three-level meta-analysis of 25 studies (50 effect sizes) was conducted to evaluate how haptic interactions in XR learning environments affect students' learning (i.e., conceptual learning, operational performance, and user experience). Funnel plots and Egger's regression method revealed no significant publication bias. Results indicated significant improvements in conceptual learning (g = 0.91, 95% CI: [0.40, 1.41]) and user experience (g = 0.48, 95% CI: [0.06, 0.90]) with haptic interactions, though effects on operational performance were non-significant. Moderator analyses explicitly identified key contextual drivers of efficacy: (1) Intervention time: Long-term (g = 0.79) and middle-term (g = 1.65) interventions outperformed short-term implementations; (2) Learning pace: System-paced (g = 0.97) outperformed self-paced; (3) Environment: Both classroom (g = 0.90) and lab (g = 0.87) settings yielded significant gains; (4) Subject domain: Health sciences (g = 0.79) and science technology (g = 0.62) yielded significant gains; (5) Haptic type: Kinesthetic feedback (g = 1.21) surpassed tactile approaches. These findings underscore haptic interactions' potential to enhance educational outcomes through multi-sensory learning, while providing actionable strategies, such as prioritizing kinesthetic tools, to optimize XR-based learning.