It was missleading, as the encoding was only for the upload stream, not the file contents when stored.