Run the same preset and settings in both engines. Both tests use WebGL2, matching procedural stress geometry, matching object/light/particle counts, explicit VR entry, equivalent orbit/pan/zoom browser controls, and an in-world headset HUD with configured vs rendered triangle counters.
Useful result: 90 FPS vs 90 FPS only means both are under the headset/runtime ceiling. Increase stress until one or both drop below the ceiling, then compare avg/p95 frame time and visual behavior.