RealityKit performance is very low

If you're planning to achieve a smooth rendering in RealityKit iOS app, 100K triangular shaded/textured polygons per scene/asset is a safe / robust maximum. Thus, to maintain an ideal 60 fps refresh rate, the amount of 100K/or less polygons will provide a rendering frame time under 16.67 ms. In RealityKit visionOS app, however, a maximum is 500K triangles. Watch the Optimize 3D assets for spatial computing (time 02:50) video which provides you with polygon count in immersive (i.e. visionOS) app.

The main question is, does the 3D model need to be prepared in any special way to work with RealityKit? The answer is evident: Yes, it definitely does. 3D asset must be prepared in a specific way: .usdz/.reality format (number of meshes in a usdz file can be around 50), poly count limit, limited quantity of PBR shaders, limited quantity of textures with base of 2 resolution, limited number of lights with shadows, etc. As we can see from your data, you have exceeded the safe number of polygons by 8.5 times. First, try reducing the number of polys of .usdz asset in Maya or Blender. Second, try disabling lights/shadows and all PBR shaders.

RealityKit's near and far clipping planes significantly impact rendering. These parameters define the range of distance (frustum volume) from the camera within which objects are visible and rendered. Objects outside this range are clipped, meaning they are not drawn, which directly affects which parts of the AR/VR scene are visible and how much processing power is needed for rendering.