Because most surfaces have cavities that might be wider below than above, so they trap a lot of light, effectively cancelling any reflection (because the light will bounce inside).
But simply using bump/normal map doesn't produce such light-trapping information, and even displacement does not. It would have to be vector-displacement, and very granular.
To create life-like digital twin of some real-world surface, the material + geometry have to fully replicate the total complexity there is. But usually the geometry in 3D is simplified, it's never as super complex and detailed, so the material needs to add that information. But the only way a material can add that remaining information, is by trickery. Thus, 100perc. PBR material will look uncanny, always smoother than should be.
And then there are special cases, like Wood. Wood has multi-directional anisotropy/SSS effect along the grain pattern, something that generalized shader cannot recreate (only true BRDF scan like ChaosScans).
So by doing some reflection mapping, you can at least partly fake it, and make it look more real.
Most people doing scanning have already realized it, it's why when you look at latest Megascans, they bake-down some cavity into Albedo, they don't diffuse it totally.
The reason why Dubcat advocated for IOR mapping, which you can now super simply achieve just by using DisneySpecular slot :- ), is that CoronaPhysicalMaterial, though almost every generalized 3D shader, just by using either normal mapping or roughness, it doesn't modulate the specularity enough. At least when you compare to real-world sample of same material. Is the Shader wrong? I guess the Devs would say no, but we already went through how many shader models and they always had something wrong :- ). To me we're still long-way from something that behaves absolutely like real material and maybe that's not even possible with generalized shader. The metals are already there, the GGX with Tail can replicate any metal almost 99perc. But the non-metals, particularly materials that have deep micro-structure, like Fabrics and Wood, those still look wrong.
And what is "Sheen" after all :- ) ? Just nice to have non-PBR fake (not needed if you use super-detailed GeoPattern for every fabric, but that would be super impractical, so hence, fake solution to rescue).
Thanks for taking your time to elaborate. I found this really interesting.
Guess I should start incorporating this into my workflow. Still got lot of questions though. You mentioned fabric and wood as materials where a mapped reflection is needed and excluded metal and more simple materials like plastic and plaster. But I guess there're more materials whose reflection we could map. Terrains and dusty surfaces usually look too shiny to me, specially in V-Ray. Not sure if that's another good case for a mapped reflection.
On the other hand, I'm not sure how a reflection map should look like for these materials. So I wonder if there's any good lecture or tutorial you could recommend.