Something similar (but not exactly the same) could be done, of course. I think of it as a Level Of Detail approach. Meaning to have the same object available with different detailness and the nearer the object is, the more detailed version is displayed (Sonic R, Andretti Racing and others).
With SGL, this can only be done on the object level, not for each individual face. Some computation - like z depth - needs thus to be done twice (LOD selection, then again inside SGL). If you would have access to the source code (a custom renderer lib), you could change texture for each face.
The different resolution versions of the same texture (most likely) need to be in VDP1 VRAM at the same time.
So the transition between low and high res textures is forced to be located at the edges between faces, it cannot be inside a single face. That is some main difference to mipmaping, I guess (also, there can be no trilinear filtering, unless you would generate those textures in software....read below).
(in theory, my texture coordinate stuff can be adapted to produce textures whose resolution is tied to the z-distance. BTW: lately I had a look on it and found a new possibility to speed up texture generation)
Well, I'd like to use LOD some day on Saturn, it makes sense to do so.
