Abstract: Multisensory XR may support informal learning for autistic adults, but sensory-friendly design often emphasizes reducing input. We conducted a mixed-methods, counterbalanced within-subjects study with 20 Chinese autistic adults who evaluated five complete XR configurations presenting the same popular-science content. Comfort differed substantially across the configurations (Friedman chi-square(4) = 46.39, p < .001, Kendall's W = .58). L3 received the lowest mean rating (2.97/5) and was tied for lowest for 16/20 participants, whereas L5 received the highest mean rating (4.43/5) and was tied for highest for 17/20. Participants described the static configurations in terms of clarity and self-pacing, L3 in terms of abruptness, simultaneous demands, insufficient warning, and limited disengagement, and L4/L5 in terms of gradual transitions, opportunities to disengage, and coordinated narrative cues. The study provides a configuration-level account of sensory comfort and identifies participant-described qualities that can inform future controlled studies of multisensory XR.