First, as far as the idea of having something like this goes, I think it's great. Particularly for large, complex, or highly formalized (in the sense of being subject to audit at least) it would be good to have this in one's bag of tricks.
Off the top of my head, here's a couple of approaches. I have not put a lot of thought into them, nor am I particularly advocating them. If all these do is spark discussion, they've completely satisfied my expectations.
If you want something uncluttered, and that easily lose its meaning and get confused with something else, you'll need to come up with a change in notation, something that's not subject to loss if the diagram properties are changed.
Perhaps you could tweak the lines of the «instance» to have a different pattern. That would be easy, and visually distinct. However, I'm not completely convinced that EA handles stereotyped connectors as well as it could. Diagrams might also look a bit too foreign.
Another approach would be to modify the ends of the connectors. Perhaps you could reuse the composition (or aggregation) notation. This would not carry exactly the same semantics as it usually does, since it is connecting a class with its instances. Another possibility is to use the notation of the nesting connector, to indicate that the instances are 'contained' by the class definition. Finally, you could use a new notation.
Fiddling around with the notation of a connector seems drastic at first, but OMG themselves has shown it is possible and usable. Consider the SPEM as an example, which Sparx has in draft. The above are considerably less drastic, since we are not modifying or replacing the underlying constraints of the UML metamodel.
What say all?
