So here is a test I ran yesterday and overnight to see how practical this process might be with a real scene. (Spent a lot of time staring at the blue spinning wheel.
)
For me I'd say this is a typical water scene, about half water and half land surface/foliage. The original water surface is a lake object and extends about 220 meters from the camera. The rendering times given are for the full 2400x1000 pixel frame, which is cropped here.
I exported two versions of the water surface, one at MPD 0.8 and another at MPD 0.6. The results for both aren't bad, though if you compare them carefully (not obvious here) there is some loss of resolution in both that affects the backlit details in the shadow areas. (Not sure why the lower MPD rendering time is slightly longer than the higher MPD time, but there it is.)
I was going to make more comparisons (MPD 0.4, etc.) but this was taking too much time and the conclusion seems obvious here.
Which is, for me, the savings in rendering time does not compensate for the time spent exporting and handling the mesh objects, which is about 25-30 minutes each on my machine. That means a real time savings of about an hour for each rendering. Again, for me, and for a scene this size, this is not significant. Plus, if you want to make *any* edits to the water surface, you have to go through the whole process again. And anyway I'd rather have the surface detail that gets lost.
So I would not recommend using the process for a scene like this. And never for a scene that is mostly water surface . . . the file sizes and processing times would go through the roof.
Maybe for scenes with smaller reflective areas, or more focused on foreground detail.
Interesting to give it a try, though.