It just will not happen. Many embedded devices build today don't even enable the IPv6 stack while it's only a configurable option away in the Linux kernel. And these devices aren't even released yet. They will after release run for a decade in the infrastructure. Sure we tell people who make them to care. Yet mostly they have more important things to care about, as for example to get them working in the first place.
An extra screen in the config box to set a static IPv6 address on an embedded device? Not seen one yet... Why? Because these embedded boxes are typically run in a seperate VLAN in the company.
For some reason you seem to think turning on IPv6 precludes running IPv4 devices or being able to reach them from IPv6. IPv6 only internal networks are still a long way away.
Corporate requirements for IPv6 are close to nonexistent, so nobody cares, nor will.
It's not that I'm against IPv6, but one has to be realistic about what to expect from the rest of the world - and a drastic change without a game-changing urgent need is not one of these things.
Lots of corporate desktops talk to machines on the outside. These machines will be a mixture of IPv4 only, IPv6 only and dual stack. The need to talk to these machines alone will result in IPv6 being deployed internally. We are starting to see ISP complaining that they can't get enough IPv4 addresses to meet their needs. It won't be long before the first forced IPv6 only sites start appearing. When that happens, offices will start to adapt by enabling IPv6 to allow the desktop machines to reach these sites.
And I'm still waiting for an example of any organization _ANY_ organization who needs to have in the order of 16 million directly communicating devices on their private network. Just a million will do as well.
Probably Google is the only organization which comes close to that order.
Pick any large ISP that manages CPE equipment. These needs to be addressed.
And even for them, there is not really an important reason why their infrastructure could not be split up between the google search cloud as one 10.x.x.x range and the gmail infrastructure as another one, for example, as direct communication between the two is probably unnecessary and managed by separate teams anyway.
Multiple routing realms are additional operational complexity. Given the choice of "deploy IPv6" or "run multiple routing realms" I'm sure many companies will pick run IPv6.
One could argue that the 'renumbering' is difficult. Yet the cluster which Google build handles server failover, swap-in and swap-out and data partitioning as one of the major features. Fail to see why they couldn't implement it on the 'private IPv4'-level either...
While I agree that on their scale, such an experiment might be valid, my guess is that it will remain as such;... an experiment with a lot of problems:
1) increased latency because of IPv6 tunneling - and Google is very latency conscious
2) less proven technology leading to exotic problems which show up even more at the Google scale - because nobody uses it
Tunnels will go away and be replaced by native connections. This is a observable trend today.
The amount of IPv6 traffic world wide is growing rapidly. The bugs in equipment will be worked out.
And for what?
To solve the 'we are too lazy to write a stupid IPv4 pool re-numbering/re-partitioning'-problem? While it can be done with a very small shell script(TM)? ;)
There are lots of reasons to deploy IPv6. We are at a tipping point where staying with IPv4 will start to get more and more expensive. IPv6 will be seen as the cheeper alternative.