Great thread, good topic.  I have been reluctant to dive into this debate until I got a bit more familiar with the advantages of tris, quads and n-gons.

If you stay in your native app, it probably doesn't matter.  XSI at render time is almost blind to these things as well as it is blind to surface mesh vs poly mesh. 

However, for mesh adjustment or sharing mesh with others who will be rigging or texturing, or for export to another app (such as Poser), then this becomes an issue.

Many apps export in tri's only so ngons get converted.  I think Marvelous Designer exports in tri's only. Hard to adjust in XSI because of the way the tools work in XSI.  Quad points are easier to select.  I think this is true of Max and Maya and Modo and any other decent modeler.

To me, the test is to go ahead and use n-gons and see if the n-gon converted correctly or introduced an abberition.  If it is a small ngon in an insignificant area, you're probably safe.

I'm working on a car and as it turns out, there are a couple of small areas best served with n-gons because they have a small area with compound curves.  To do it in tris would require a lot of tris.  I have no idea how the final render engine in, say, Poser, will react to those until I test.  Mental Ray is fine with them.

Personally, I think it has a lot to do with playing nicely in the sandbox with others.  If others need that orderly quad to access for rigging and relaxing during UV, then so be it. 

So I think it depends on your pipeline and final user.