How hard would it be to implement varying resolution across the monitor with this approach, e.g. to combine with an eye-tracker to ensure that only the parts that the user looks at, are rendered in full resolution?
Varying resolution is trivial with raytracing, however a big part of the work here is not tied to screen resolution. Lighting is sampled far below screen resolution and needs to be reconstructed over multiple frames. Eye position adds no useful information here. Doing full-resolution "cinematic quality" path-tracing would require hardware that is magnitudes faster.
Having said that, variable shading rate is a feature of the newest NVIDIA GPUs (and supported by Wolfenstein: New Order) and it could be used with a sufficiently low-latency eye tracker rather trivially as well.
