I found the Babel Experiment described here very interesting.  It claims to show that if you only allow people to use UML and not to talk, you get a better result than if you allow them to talk.  The better result is only that "their diagrams were more detailed, more elaborated and elegant", but still it is an interesting result.  I did recoil from the conclusion that "UML Metamodel is a great example of good design" though - this experiment has nothing to do with the UML metamodel, which is the total opposite of a great example of good design.