If that is true then it is not a very smart design of 3DGS. DDS should not be internally decompressed, it should be sent in compressed state to the graphics card.

What I am talking about is indeed a good testbed for checking:
- how big can textures be
- what format is best
- is it better to have little small vs. a few big textures (containing other textures via atlas)
- is it better to have bigger poly models vs. many with small poly budget
- how many texture memory can be used (with or without lightmaps)
- how much does it help to remove non-visible polygons
- how big is the impact of several shaders
- how big is the impact of animated models
- ...

All this can be seen best with a sample scene containing everything in the aimed quality. And if someone dont has the models and textures, it is easy and really cheap to get them from decent indie-friendly shops