I think this is a kind of a milestone in 3DGS.
Excuse me but I don't follow: 5*3 (textures) + 5*2 (normal maps) = 25. What about the remaining 3 channels? Are they used for light mapping? Or you can use them as you like, including nothing?
As for optimization:
1. I'd only use normal mapping with some textures (small ones, if I understand correctly from HeelX that the most intensive texture "processing" is done per texel).
2. I'm saying this because I use differently sized textures for each terrain type. For grass, for example, which I think it has to be more crispy, I use a larger size, small scaling or both. And I spend a lot of time weighting the appearance/sizes balance.
3. Now it's unclear for me whether the whole terrain gets rendered (texture-wise), even if the camera only sees a corner, or it's clip space optimized (I have to dwell on the pipelines study). Cause if it were optimized, I'd consider (a little) more processing friendly to use smaller scaled textures than upscaled ones at the same texture size.