If you are looking for best-lighting, and no discontinuities, you wouldn’t want to avoid using texture bombing on your roughness, etc if you are using it on your basecolor as well. Those textures all line up to provide the best-quality image. Unless you are using a solid # for your roughness, etc, TBing basecolor but not roughness et-al would cause some visual artifacts.
Ideally, you want some UV math you can share between multiple samplers vs having to run multiple-samples across basecolor, roughness, etc. This would recycle the math of the UVs across multiple endpoints and amortize the overall cost.
At the very least be packing your greyscale textures into a single texture to save on samplers/samples.
One thing you CAN do to claw-back some performance, whatever you end up doing is look at using say, full-scale basecolor but a mipped-down roughness, gloss/specular, normal, etc. Instead of 4k/2k for all the textures use a 4k for basecolor + roughness (RGBA) (roughness is most important besides color, IMHO) and a 2k Gloss/Spec+2channel normal+ AO, or even 1k. At least you can save on memory for less-important information.