The two eye views are almost the same image. How do you avoid paying for the scene twice?
Only the view and projection matrices differ between eyes, so scene traversal, culling, animation and shadow maps can be done once, and multiview rendering lets one draw call emit both views with a per-view matrix chosen in the vertex shader. What cannot be shared is anything computed in screen space.