Right - Travis' original concern was T junctions where smoothing is needed - basically an app that does per-pixel lighting and smooth normals _will_ show a nasty artifact at T junctions that were smoothed. So that's a big one.
For flat shaded intersections, there's no lighting advantage, and the artifact is 'cracking' - if you don't see it now in today's programs, you probably never will; the fact that lines get drawn on top of the seams tends to hide some sins.
The problem with pre-processing I think is that you'll induce vertices not just for T junctions, but for the -potential- for T junctions for other parts...see the example above of the torus section; it will become more vertex intensive for everyone who uses it, just to fix a problem with one minifig face.
For flat shaded intersections, there's no lighting advantage, and the artifact is 'cracking' - if you don't see it now in today's programs, you probably never will; the fact that lines get drawn on top of the seams tends to hide some sins.
The problem with pre-processing I think is that you'll induce vertices not just for T junctions, but for the -potential- for T junctions for other parts...see the example above of the torus section; it will become more vertex intensive for everyone who uses it, just to fix a problem with one minifig face.