I completely agree with you. By creating overlays I propose to do exactly what you are saying: constrain the topology only to the interested parties. This puts away the scalability concerns but raises the question how bad it is in terms of configuration effort, specifically, how difficult it is to add a new PE (a member of a subset of VPNs). I know that BGP configuration is not exactly automatic either. So how big is the difference in the configuration efforts?
I know for a fact that some L1 network operators are interested in L1 VPN application but reluctant to deploy BGP. They do deploy for quite some time already GMPLS based control plane with OSPF-TE. Why in your opinion the multi-instance OSPF solution will not be acceptible for them?