Are you talking about having a VPN terminate on a " virtual" interface that relies on end to end connectivity rather than connectivity on an individual interface?
There are ways to do this, but the solutions that I imagine of have nothing to do with a loopback address. In the context of security, zones are probably more appropriate. If you have two interfaces that have separate internet/WAN connections, they would belong to the same zone. Policies are applied at the zone level, not the interface level. In this case, the VPN connection would be created in that zone using whatever means necessary to make the connection, either or both interfaces. In this case, I guess you could use a loopback address, though for the purpose of identification, rather than an actual functioning IP address. This is a very simple explanation and leaves a lot of routing out, but that' s somewhat what I think of. The Netscreens are much further along in this than Fortinet.
Hopefully v3.0 will resolve some of this. I don' t expect them to make such drastic changes in single release though. Not only would the code have to be dramatically altered from what they currently have, most firewall technicians would be lost because the concepts in VPN would be so dramatically different, albeit, that' s the way it shoud have been done in the first place. There are many changes I would like to see in VPNs to increase their flexability.
Is this what you are talking about?