Immersive VR and MR exergames are increasingly used to support physical activity and balance training in older adults, yet evidence is often based on single-session or very short-term evaluations. Because initial responses can be inflated by novelty, it remains unclear whether greater immersion or technology mode meaningfully supports engagement across repeated sessions. We conducted a 6-day, between-subjects study with 35 community-dwelling older adults (71.4% female; mean age = 69.0 ± 8.0 years) who played a Tai Chi–themed balance exergame (6 sessions over 6 days; ~ 15 min per session) in either Virtual Reality (VR) or Mixed Reality (MR). We assessed user experience using the GEQ and cybersickness symptoms (VRSQ) and logged in-game performance. Compared with MR, VR produced significantly higher Sensory and Imaginative Immersion (p < .001) and Flow (p < .001) compared to MR. However, this extra immersion did not translate into superior outcomes on final positive affect (p = .440) or cybersickness symptoms (VRSQ scores, p = 1.000). A multiple regression analysis, which controlled for multicollinearity by excluding redundant predictors, indicated that competence and challenge rather than technology mode, immersion scores, or objective performance were the only independent predictors of final positive affect. These results suggest an “immersion paradox” within this specific context: maximizing immersion alone appears to be insufficient for sustaining positive affect in older adults. Design should prioritize challenge–skill balance and dynamic difficulty to maintain engagement over time. Cybersickness symptoms (VRSQ) did not differ significantly between modes; therefore, any potential safety advantage of MR should be treated as a design hypothesis for future work rather than a conclusion from the present data.