If you want to display some text in a WebXR scene, you don’t need to create a 3D geometry in the first place, did you check out this example ? This would reduce the number of polygons and help a bit, but it still produces meshes, and creating meshes every frame always lead to performance issues.
The best thing would be to pre-create the characters meshes, and only show/hide them with their visible property and position them to create new words.
This will help, but if your text is big, it’s still a lot of independent meshes, which is still a problem for performance. The solution is to use InstancedMesh, so you only have one mesh per character, visually duplicated by the GPU.