I guess we see a bit of light coming from the top and sides that would normally be hidden, but that ray tracing is making the renderer very sad because sometimes you sample from the back of the polygon that doesn't have texture?
Or something like that, there's definitely something funky going on with the view of the top surface that's unrelated to GR.
Its hard to explain exactly what's accurate here when you're looking at the precise shapes - part of the problem is that this rendering is extraordinarily complex to do in real time, and it relies on the size of the object (tris) being 'small' relative to the curvature of spacetime
There's actually no texturing here at all though, if you see a face disappear its because general relativity is hard. The extra disconnected pieces are something to entirely ignore as well, and occur because tris become nearly parallel with rays leading to long intersection lengths (as well as numerical issues), leading to the local flatness assumption breaking down
Or something like that, there's definitely something funky going on with the view of the top surface that's unrelated to GR.