The fixed size is quite okay to work with, but I think the node should be doing some internal auto-padding. Right now it overwrites the shared memory only with as much bytes as are inputted and leaves the rest as it is.
One patch-around is to use Pad (String) but I don’t think that is the smart choice for most cases involving Raw. The other is to + (Raw) with a long-enough bytestream, which is quite cryptic to read.
in any case i’d argue you’d better reserve the first 1-4 bytes to encode the length of the current data and better rely on that than to look for some padding at the end.