If MetaNet provides a single-subnet, multi-access shared medium abstraction, then a link local addressing scheme might be good for something. Bill Manning once popularised a scheme for using link-local addressing at exchange points in order to avoid RIR policy, to the point of running code even, but it was not widely adopted for reasons that escape me.
We tried using link locals as addresses to peer with each other. The problem with link locals is that you have to instruct all your software as to which interface a particular link local address is on. I'm not sure how well BGP's next-hop calculations work in this case, at one stage we were getting the next-hop being set to the peer always. Generally link locals seem to be more trouble than they're worth for setting up permenant routing infrastructures. Cheers James