I think I now get what Justin's saying. The problem with Wilhelmine Germany he mentions–that the democratic body didn't have control over foreign policy–I think points to the kitten/wolverine problem. To say that a regime is a democracy is not to say much. Democracy, per se, is certainly not a very useful category for social scientific generalization. As I mentioned a couple posts down, there can be immense variation in institutional structure within democracies, and the specific structure probably does much more work than the generic type.
My guess is that democratic peace theory is a bit retarded because its too atheoretical and there is not sufficient attention to institutional dynamics. I'd like to see a good NIE/public choice comparative analytic narrative of an instance where democratic institutions seems to have prevented war and an instance where they seem not to have. (Dissertation topics are free here at the Fly Bottle.)
My totally uncoached guess about why democracies don't attack each other much is that (1) democracies in general make less agressive offensive warlike noises, (2) democracies recognize each other as having a kind of legitimacy.
I know next to nothing about this, but I won't let that stop me!