In talking with Jason my impression is that he wants to implement this shift via smaller changes like TS-4092. My concern is if we don't have a final structure to which we are moving we could end up in a worse mess than we are now. And if we are going to make significant changes I think in for a penny, in for pound. If we are looking long term I think the overall pain is less if we pick the best structure we can and move toward that, it will be better than having to deal to do semi-major changes repeatedly and much better than doing it piecemeal without an overall goal.
I agree with Jason that we need to think seriously about testing and how we can arrange things to make that easier both in implementation and having people understand where to put things. That, IMHO, is a big mess in addition to the ones Ming points out. There are several significant architectural changes on the horizon that are going to require far more serious testing than we have in place now.
