So as I am toying around with lighting shaders, great looking results can be achieved. However, I struggle to fully grasp the idea behind it. Namely, the microfacet BRDF doesn't line up with how I intuitively understand the process. Expectedly, the perceived brightness on a surface is highest at NdotH, but this gets to be increased two-fold by the denominator as the L and V angles diverge. The implicit geometry term would cancel this out, but something like Smith-Schlick with a low roughness input would not do much in that department, making gracing angles very bright despite there being no fresnel involved. The multiplication of the whole BRDF with NdotL then only partially cancels it out. Am I missing something, or should a relatively smooth metallic surface indeed have brighter highlights when staring at it with a punctual light near the horizon of said surface?

This presentation (and accompanying paper) talks about the cosine/dot products in the denominator a bit, amongst other things.The reason for these terms is that the BRDF always deals with with a surface patch whose area is exactly 1, but the projected area from the eye's point of view and the light's point of view is not 1 (they're proportional to the cosine of the angle between the eye/light and the surface normal). These terms data back to 1967(!), when they were discussed in Torrance and Sparrow's paper about off-peak specular reflections.