Almost more interesting is consider the "financial interchange network" where 100 companies come together each with 20 networks. I believe even with that relatively small number of networks (2000 total) the probability of collision is well more than 50%.
My company already does this (financial interchange network) using registered globally unique IPv4 addresses. When we transition to IPv6 I see no reason why we would not continue using globally registered IP addresses. The routability of these addresses is not relevant to us because our network is disjoint from the Internet, therefore the addresses only need to be routable on our network. Other organizations have similar types of international IP networks that are disjoint from the Internet. A competitor of ours, SWIFT, operates their own IP network and probably interconnects many of our customers as well. In the automotive industry in Europe there is the ENX http://www.enxo.com that is a similar disjoint network. When I was last involved with ENX it used existing IP backbones but traffic was exchanged over special dedicated peering connections with an enforceable SLA between the two peers monitored by the ENXO. I suppose that SITA must also run IP on their network these days. In all of these cases, companies do not just "come together". The interconnection is carefully planned and often involves some sort of partitioning of the companies' internal networks. For instance, many of our customers connect their trading floors to our network however those trading floors do not have direct connectivity with the Internet even though all of our customers do have Internet connectivity. We are now getting into a very complex area of enterprise network architecture which it is pointless to second guess, because somebody, somewhere will implement any architecture that you can imagine. The point is that when companies interconnect networks, they do not simply merge their internal networks with the internetwork. With our customers, the interconnect is done for a very specific defined purpose and engineers on both sides ensure that the interconnect does not carry traffic that is out of scope. In the case of corporate merger, there would likely be a network audit (due diligence) followed by a decision as to how to deal with the small number of collisions, either renumber them or install an IPv6 NAT device. The decision won't be made based on elegance or RFC compliance, rather it will be made based on the business case. I suspect that if you do the actual maths for companies the size of Microsoft and Time Warner, you will find that the number of collisions will be so low that the business case will lead to renumbering. If anyone disagrees the I would really like to see a worked example of the math.
Best I can tell renumbering will be NO LESS expensive in IPv6, and actually will be more expensive, since the IPv6 proponents seem bent on requiring everyone to have 6 addresses to do anything useful, so it will be address management effort x6.
Address management and renumbering are so important to today's enterprises that they all use some toolset involving LDAP directories, databases and DHCP servers. Multiplying the data content of the database times 6 does not necessarily change the level of effort. I'm surprised that no-one has pointed out how easy it is to get an IPv6 allocation today and set up a registry for globally unique IPv6 addresses for private networks. The initial allocations are large enough to not have to ever return to the RIR for more, and if a company can sell people on the idea of allocating globally unique IPv6 private addresses in the same way as IPv4 (i.e. /128 per host rather than /64) then it would last a long time. By the way, there is already the so-called deprecated site-local address range that folks can use. I assume that "deprecated" just means that you need to explicitly configure the ACLs at your border rather than relying on some automatic routing behavior. Perhaps the draft could be recast as a "best practice" for using site-local addresses via a hashing algorithm that makes collisions less likely. --Michael Dillon