Still today, I often find that the notion of DPI does not work (for screens at least). The first example that comes to my mind is whenever I plug a new screen for, say, a multi-head setup.
Let's assume that all is well: we have both LCD screens, the native screen resolution is detected upon new connections, as is the actual physical size. This is unfortunately not good enough: it could very well be that the two screens are not seen from the same distance, or even by the same person in the first place.
It's not really that the concept is flawed, it's more like it misses at least another factor. Enforcing a fixed value will not cure the problem, since your second screen may as well be very big these days: while fonts may be good on your laptop at 96 DPI, you may want your assistance to be able to read from 10 meters away on your beamer...
Therefore, I suggest that we also define DPI with the notion of depth from now on. I have no idea if this idea is new, or already patented or whatever. I'm pretty sure it isn't (new). I will refer to it as « DPI 3D » (tm)(Copyright © 2009 Me) in the future. Who knows, maybe you will too :)
Seriously: I'll welcome comments and suggestions about how to solve this, and whether my idea is worth propagating. Have fun!
I said I'd never blog
DeforaOS, NetBSD, reverse-engineering and stuff



