They don't. I never said that they're using SSR for the mirrors (I explicitly said that they don't).
Ah okay, it's just that your phrasing let me believe you were under the impression that I thought they were using SSR.
What I think is going on is that every frame, they render a cubemap with the camera placed at the center of the car. The cubemap is then used as, well, a cubemap. It's not precise at the mirrors because it was rendered at the middle of the car. So in that particular shot, instead of the mirrors showing what was behind and to the left of the mirror (the body of the car), it shows what was behind and to the left of the car (the bridge railing).
My tiny brain can't see how the scene would be rendered onto a cube, here.
But if I'm nevertheless following your thoughts, you think that they're rendering a single scene with a fixed camera on the 'roof' of the car pointing straight behind, and cut/crop this render into as many pieces as they need to match the cars' mirrors, right?
Let's see if we can check that with an example from gameplay...
So here's what you're supposed to see when looking back:
And here's what's rendered in the cockpit view:
The McLaren F1 LM is an interesting car, because it has two inside mirrors.
Quick and dirty try to get the perspective right, for comparison:
So, yeah, it looks like a single render split into the two mirrors.