Despite my agreement with Matthew's "Fat Models are Good", he starts to bump up against another point that I run into regularly:

Less Code is Better

End of Story.  There are no exceptions.

Sure, lots of people want their code to be extensible, flexible, powerful, handle 100% of the cases flawlessly, handle all input, fail gracefully, be readable, blah, blah, blah.  All of those are great goals but Less Code (that still meets the specs) is still better.