Zfs dedupe has memory overhead for writes. Technically the read overhead is zero since the metadata on disk just points to the correct file chunks.