I didn't RTFA but would point out that the performance problems I've run into with IPv6 aren't related to the specification itself but buggy implementation code from vendors, which I expect to improve with time.
I thought about using an additional ULA subnet (ipv6 supports multiple subnets), but apparently many vendors do not yet support RFC 6724, or support it poorly with regard to ULA addresses. As such, I get mixed results with devices trying to use the ULA address for internet egress.
NAT66 (npt) is another possibility, and some people end up having to use this when they have more than one ISP (multi-wan balancing) anyway, but I haven't found a good way to update the nat target when the prefix changes yet (might try moving to openbsd and try ifstated calling a script to a pf table or something).
Part of the point of IPv6 is that your addresses reflect the real network topology - when your upstream changes, your addresses change. The right thing is to ensure that your router is also updating DNS for your internal devices, and use names to refer to them. You shouldn't expect anything to have a truly static IPv6 address.
> The right thing is to ensure that your router is also
> updating DNS for your internal devices, and use names
> to refer to them.
For most devices, a constantly changing IP is well and good (eg ipv6 privacy extensions). However, for servers it is less than desirable.
The only solution I've heard so far is to have a client on each machine that would update the DNS server with the new address, which sounds terrible.
I'm just about at the point that I'm just going to install DHCPv6, but last I heard not every OS supported it?