I agree that we are talking about incomplete documentation and this is one of the biggest problems with translated documents. People have a tendency of only having the bits they need translated. This leaves you with partial translations. At the moment we use backout languages to find the most appropriate language but which raised the original question.
One of the functions the system has to be able to do is report in translation coverage (ie how many documents are in all/some/one language). Your first markup makes it possible to run coverage reports that would show that some translations are missing. However the second is a lot clearer and would allow easy identification of exactly what part of the translation is missing.
Thanks for the feedback
UnderMine