Did you enable dedup on your dataset? Read
this. Pay special attention to the note under Table 6.3a.
Consider that:
You very likely don't have the RAM to handle deduplication.
You very likely won't have enough duplicated blocks to actually see any benefit from deduplication. (What's your use case for this?)
You'd very likely see much more benefit from simply enabling compression. (Again, what's your use case? Even compression is likely best enabled only on specific datasets.)