Actually, Smoth, that was a pretty good summary
Basically, he's entirely right- never use multiple faces to do a flat surface, unless you have an excellent reason. And, if you
must do that... then use seperate squares/rectangles/whatever, don't subdivide the surface... and move the seperate squares slightly away from the main surface. Viewers will never see it, if you keep the distance small, and if it's far enough to prevent clipping problems, then it's a great technique for things that are using a repeating cubemap with windows, etc., etc.
Why do it that way? Why not just cut the face up? In a word:
welding.
Welding is the working artist's term for a couple of different techniques used by 3D modeling software to blur surface normals between angular facets. It's a way to cheaply create the impression of curvature, basically, by blurring where the sharp edges of joins occur.
The most modern, cutting-edge game engines are moving away from welding, in favor of another (more complicated) technique called
normal mapping, where an image map is projected onto a low-poly version of a high-poly model, allowing the rendering engine to interpolate the resulting light values with the normal map, creating the illusion of curvature on a surface that has a much lower real tricount.
These days, considering just how many polygons modern video cards can handle without significant performance problems, it seems like normal mapping is a cutting-edge technique that has arrived a little too late to be useful for anything but movie-makers and other super-high-end content producers (where total polycounts in scenes still matter a great deal, because then we're talking millions of tris), but I'm really not qualified to talk about this in any great detail, so I should shut up before I start sounding as ignorant as I really am
With buildings, especially things that we're trying to get "to scale", our objective, as modelers, is to pack lots of detail into rather small things. One of the best ways to save our texture space, but still have very high details, is to re-use texture sections over and over again. We're always tempted to do this by cutting up our walls into squares of equal sizes (or, at least, keep them square/rectangular) but if you're going to do that, you need to un-weld the vertices of your model prior to importing it into Spring...
I have no idea how much (or if) UpSpring or Spring automatically welds things, but it appears to do so. I meant to ask Zaphod about that at some point, but I've wasted enough of his free time with my other questions and newbie mistakes I'm not real keen on making him stare at some sourcecode from the core renderer
