In case you have known me or this web log for a while you probably know that I am all into the evolutionary style of software development. In other words, I am a strong believer in the theory that software design - and software architecture as well - should be tackled step by step, from very easy to (maybe) a little bit more complex, from one requirement to the next. My experience so far has been that almost anything which had to be designed or "architected" could be - or could have been - done in the evolutionary style; see my talk on Evolutionary Software Architecture for more details on that subject.
I emphasized the "almost" in the section above not to stress the impossibility of evolving certain architectural characteristics - I just haven't seen it being accomplished so far - but to point you to an analogous theme: Biological Evolution vs. Creationism. It's the likeness of rationale that strikes me: Creationists very often take an isolated biological item - say the third vertebra of a frog's cervical spine - and argue that since no biologist has shown how this bone could have developed in many iterative steps it must have been created by God. As soon as this issue has been tackled by a PhD thesis or five the next detail will be used to "prove" that "intelligent design" must be at work (see Richard Dawkin's book The God Delusion for more about this topic). I've had to face many similar arguments from proponents of up-front-design and -architecture: "How can security, internationalization, persistence etc. be built into a system step by step and in retrospective? You have to do it right from the start!" I often don't have an answer then ... until someone describes how she did exactly that. I've seen that happen with internationalization, transactions and lots of other stuff and I'm sure more seemingly uncrackable nuts will follow.
Analogies of this kind are a slippery slope, and I don't want to imply that proponents of up-front-architecture have as weak a case as creationists do. No, I'm fully aware that there are systems that have been developed that way and that are as healthy as their evolutionary counterparts. The thing I want to stress, though, is that a few unsolved sample problems are not enough to weaken a theory that comes with strong supporting facts.