As it stands, there is actually no reason to turn on dedup (apart from possibly less I/O) as you can’t actually use the space you save.

In short, they created the dedup engine and did all the work to make dedup possible in the filesystem, but never updated the dataset model to allow people to use the blocks freed by dedup’ing? Wait, what? They are aware of the problem, and I assume they will fix it, but still. This is a little crazy.