I seriously question the wisdom of SunRPC. We want a situation where we can
predict boundaries so that we can reliably read blocks or share them into
the disk caches. Thus knowing the format is always read 128 bytes of header,
process, think, read data block direct to disk cache. Rather than piecemeal
scanning of headers to get to the data.
Sure it makes the protocol suck rocks over slow links, but on a slow link we
can afford the compute time to compress/decompress it at another layer.