I saw Grady Booch give a presentation at SIGSCE one year -- actually a very interesting guy and a dynamic speaker. He said a quote that has stuck with me since: "Hardware eventually fails. Software eventually works."
Grady's a smart guy but UML has been an utter failure.
Theres often more interesting stuff to be said about why something failed than about a language that is mainstream designed by committee and compromise.
I don't think it's fair to say UML itself failed. We did get to a point where to see a project modeled in UML gives you a strong desire to turn and run (and this is usually the correct thing to do), but it's more to do with abuse and stupid users who don't even make the difference between modeling the behavior and modeling the implementation.
Don't get me wrong, UML needs some serious change. However, there is some value in the process and thought that goes into creating the UML diagrams for projects, especially larger, more complex projects. I find that it helps me better visualize what I'm making and spot problem areas before they become real problems.