History shows people will use private address space for a variety of reasons. Getting rid of a published range for that purpose will only mean they use whatever random numbers they can find. This has also been shown to create operational problems, so we need to give them the tool they want to use in a way we can contain the fallout. Site local is defined to do that job, and we do not have WG concensus on depricating it.
As I wrote previously, one must understand the history in order to understand its applicability to the future. The reason there was a problem at all was that there was a not just a non-zero chance of an address clash but that this percentage rose based based on the class of address chosen. If you took 80.1.4.6 you clashed with all of 80 because classless addressing had not made it into the base yet. The chance of this sort of clash in IPv6 is miniscule. Not that it's a good idea.
Even so, if you used upper range class B or C address space the chance of a clash at the time was low (still is). And indeed in the case of a merger you were better off than using net 10 because the likelihood of clash with net 10 is high.
Eliot
