One additional comment...
Aside from CIX/NSFNET routing problems, we have seen some of the
consequences of multiple backbones with the current T1 and T3 NSFNET
backbones and the interconnect between the two. That's how we find
routes from the East Coast to the West Coast and back again, especially
when the active interconnect is at SDSC. The routing arbitrator in
this case is the same as the operator of both backbones but that
doesn't help since the problems are fundamental to basic routing