> Once you have a conceptual model with nice properties, the
> implementation is just a "technical exercise", and it can of 
> course be done in different ways. Typically you encounter the 
> same space-performance trade-off as in nearly all the other 
> kind of implementations. As an example, just consider traits:

Sure, but from my perspective, the VM is currently implementing a rigid
conceptual model.  If we can get that out of the VM, or at least make it
cleaner to override that behavior, then it's much easier to play around
with other concepts in behavior composition...and that's what I'm after.

With regard to how behavior is bound to a given message, perhaps the
only assumption that we should make (in the VM) is that we can make no
assumptions.

- Stephen