First off, I agree completely–I really don’t recommend using the ZFS primitives to manage Samba.
Second, I can tell you a potential benefit: if the management is implemented within ZFS, ZFS can store configuration information about your shares in custom properties embedded directly in the datasets being shared, potentially enabling you to have them automatically shared again on other systems you replicate to / restoring their sharing on those systems using the embedded properties / etc.
Third, sorry, I can’t believe I didn’t see this topic from so long ago! <3