> Initially, we implemented this solution using a bytea column, but we found that the retrieval performance was unacceptable. It required the whole file to be r

> Initially, we implemented this solution using a bytea column, but we found that the retrieval performance was unacceptable. It required the whole file to be read into memory before being sent back to the user (as opposed to streaming the content in smaller chunks). How large are the objects on average?