This is essentially the old argument over whether it's better to patch old bad code, or whether it's better to rewrite old bad code. I think all of the same responses would apply, i.e., whether or not new problems might be preferable to the existing ones. Since much smarter people than I have written carefully-considered essays on that argument (and without coming to a clear conclusion), I'll just let them speak for me. :-)

It is OK to have an in-context opinion, though. My opinion on this is to change as little as possible, and one thing at a time.

Let us not forget the object lesson from Digg's big redesign.