I've been using 2.7 for few months now and I'm glad it finally goes official. The concurrent serving of requests really does make a difference: for app that is mostly I/O bound and isn't using a lot of asynchronous RPC, it greatly increases request throughput. For us it not only nullified the pricing change from last Fall, but actually reduced the overall cost, compared to Python 2.5 + previous billing.
Now when it's official, GAE looks like a much more viable and cost-effective platform than it was for the few passing months.
You can get IPv4 addresses; the going rate is about $1/month/address. Whether Google's infrastructure can handle zillions of custom IPs is a different story.
being able to finally use PIL and lxml on appengine is huge, and 2.7 also goes some way to resolving the billing issues. 2.7 going gold is starting to sway me back towards using appengine again.
I'm glad it's gold so I can use more modern libraries again. But the reduced cost thing is not a guarantee. If you do a lot of rpc calls per request, multithreading actually makes latency and instance count absolutely horrible.
Of course this seems kind of ridiculous to me, if you aren't using the rpc's what the heck is your app actually doing?
So like everything, you need to test your application to see what configuration is right for you. For the app I maintain, python2.7 single threaded is only slightly slower than 2.5 but I expect that to improve over time and it was worth switching now just to not have to write for python2.5 anymore.
Costs are higher, but still less than leasing a data center and employing staff to maintain it.
Now when it's official, GAE looks like a much more viable and cost-effective platform than it was for the few passing months.