Feedback-based iterative refinement is important in the development of any human-computer interface. The present work aims to evaluate and iteratively refine an immersive learning environment called Scale Worlds (SW), delivered via a head-mounted display (HMD). SW is a virtual learning environment encompassing scientific entities of a wide range of sizes that enables students an embodied experience while learning size and scale. Five usability experts performed think aloud while carrying out four interactive tasks in SW and compared three different design options during A/B testing. Improvement features based on the feedback from an earlier SW usability evaluation as well as HMD-specific features were examined. Usability experts completed the post-study system usability questionnaire, the NASA task load index, and a bipolar laddering survey that collected subjective perception of specific SW features. Results show that the progress panel (an improvement feature) was informative while the instructions (another improvement feature) caused clutter. The experts indicated clear usability preferences during A/B testing, which helped resolve three sets of theory-usability conflicts. The overall assessment of SW paved a path for theory-usability balance and provided valuable insights for designing and evaluating usability in immersive virtual learning environments.