This is not possible. The encoding for blobs in the DB is not documented. (at least it wasn't when I looked into this in 2010) You can either migrate to a file based binary store or use something like the rest api to retrieve them.
first, your answer's in Extracting binary (user uploaded) document directly from the database something about removing the first 30 bytes of data which contains the filename or something.
but how did you link files bigger than 8MB? relating the body or attachment to a file is good, but after 8MB Jive starts chunking and things get messy. Did you figure out how the chunking works at all?