Just a couple thoughts on that...
I think we are going to aggregate our fact data to not having such complex expressions which are kind of hard to develop and debug. This will make the qvw files bigger since basically the same data are in their twice but makes it easier to maintain and develop.
What's your thoughts on that?
I also play with the idea to spread them out into serveral qvw files and link the files so the fact tables can be designed for specific measures. A user would not notice this. He just jumps from one qvw file to another but each of them has a separate data model.