There are two mechanisms for this.

You can run DHCPv6 and have it hand out info but not addresses via a DHCPINFORM. This also works in IPv4 also but not many know about it or use it. In a nutshell you setup a subnet but don't include a range of IPs to hand out. You simply setup DNS servers and maybe a DNS domain name, ntp, and whatnot. The clients will autoconfig but also run a dhcp client to get the DNS servers defined.

The other (and better IMHO) method is that you can include RDNSS info in the router advertisements. So for autoconfig to work you have to at least advertise the subnet and prefix that clients should use to form a complete address during autoconfig. The RDNSS (recursive DNS server) advertisements are picked up and used by the client as DNS servers. This method has less adoption but I think this is ultimately going to be the preferred method once it's supported more widely. See the radvd.conf man page for more info.

The router advertising is a part of IPv6 that is poorly understood or completely unknown to many people but they put some pretty good though into it. There is actually a mechanism to renumber an entire network using primarily router advertisements which is pretty cool.

There are a few things that you don't understand.

If you get a /48 from your ISP (standard allocation recommendation by IANA and by existing v6 practice) you can run as many /64 VLANs as you want. You aren't forced to run a single VLAN when you run IPv6. You can still subnet six ways from Sunday. Your comment talks about a single subnet but that isn't the norm for IPv6 deployment.

You don't need to convert everything all at once. Experiment first, then roll it out on a DNS server or a mail server.

You will have to maintain two sets of addresses for the foreseeable future. So does everybody else. You can stay on IPv4 but at some point you will need to connect to somebody who can only get IPv6 addresses. That might be 3 years from now or 10 years from now but this is going to happen. IPv4 will be exhausted - this is a fact that a lot of people are having trouble dealing with but it doesn't have to big bad and scary. IPv6 isn't really that different from v4. They both pretty much do the same job - yes there are differences but once you work with for a short time it's not rocket science - it's just basic networking.

The weak part of IPv6 is ISP delivery. There is a dearth of providers who are providing dual stack to all of their customers and this is right now the biggest barrier to rapid adoption, particularly in the North American market. This is going to change pretty rapidly over the next 2 years and alredy has in other regions.

I agree that IPv6 is scary but a true geek should see this as a learning opportunity rather than a departure from a comfort zone. IT people are supposed to be ahead of the curve. Yeah - maybe you don't roll out IPv6 until their is a solid business case for deployment but there is a business case now for experimentation so that it won't be a fire drill when it comes time to deploy because of an actual business requirement.

This demo text client is pretty spartan. I can't wait to get my hands on the HTML5 client. I was able to get this running on Debian Lenny on EC2 pretty quickly. I got two instances to talk to each other across the Amazon net and I could invite people from the 2nd instance to participate on the 1st instance's waves. So the fundamental server stuff seems to be working. Has anybody tried this out with ejabberd?

Why wouldn't you just use the timestamps and let the most recent win. Even if an offline client had a bad timestamp and something got stomped you could just replay back to a "good" state and copy paste your way back to normal. Anyway - there are smarter minds than I working on this at the wave federation protocol site.

Or on a large VM cluster - which thousands of data centers have in production now.

Japan started fingerprinting inbound aliens last year.

The thing is - Windows Vista and Windows 7 aren't really a fundamental departure from the past. For example, I applaud Microsoft for finally getting on board the IPv6 train with Vista and Win2K8 but what happened to rewriting system services and the Windows shell in managed code (.NET)? That would be a fundamental change that would justify a compat VM container. Microsoft is really giving customers the worst of both worlds. Making only incremental improvements to their mainline OS's while creating a backwards compatible VM which is simply more cruft to throw on top of an ever expanding pile of backwards compatible cruft.

The Cisco blades are going to support 192GB of RAM which is one of the most important constraints on a VM host box. For the density and form factor this is better than the competition at the moment but I have no doubt HP, IBM, and Dell will respond quickly to keep their customers in line.

Amazon S3 has the ability to do this. The issue is you pay per request so it could add up if you do something significant with it.