It would seem off heap messages could be a base for this. I've not gone all the way into the weeds on those, but it seems like it could be possible to start with sending a whole message to another process by bumping a refcount and adding a reference to that process's mailbox without needing to make a new copy.

If that works, and is useful, you could extend by bumping the refcount, and adding a new type of subreference to a process's mailbox.

I don't think you'd want to send something referencing multiple off-heap messages or adding some terms and a reference (although maybe, who knows), because that would be a lot of complexity. And you would also want to be sensitive to subreference memory leaks, like can happen with binaries where a binary is large, but only small subreferences stick around, leading to lots of wasted space (this is extra fun when you put a small subbinary into ets/mnesia and it lives forever (well until you restart beam, but that can be a long time).