Having only one window manager implementation does not mean that it may only have one behavior. For example kwin does already support traditional floating windows and tiling. With a somewhat sane architecture the code may still be readable and maintainable even with several more different ways to manage windows. And I don't see why it would be more difficult to experiment with new ways.
Things we would lose are having extremely lightweight window managers and the choice about the programming language it's implemented in. Personally I don't think that the first one is really a loss. Window management is no performance critical task and I guess that even a completely new behavior would be just a couple of 100 lines of code. A device which is able to handle a graphical UI would already have enough memory that this doesn't matter at all.