"... It wouldn't be pleasant to spend time to write an app, become popular (and everyone hopes to become popular), and then have Google turn around and say, 'Sorry, you blew out of your free quotas, and we aren't ready to bill you to extend them.' Just as bad: 'Your free app is popular, congratulations! You owe us $10K for the bandwidth and storage you used last month.' ..."
At the same time the entry level cost for exploring an idea in code has been reduced yet again. Explore an idea, create something and see what happens. If it works then inject some funds. If it fails, the outlay is time and effort only.
Except there is no way to inject funds to the idea. There's no way to pay Google for more cpu/bandwidth/storage. The system is also very propriety and there is no (easy) way move an appengine app to a new server.
I'd much rather get some cheap hosting somewhere else than develop for free on appengine. Its an interesting platform but there are still too many limitations for me to consider it for serious use.
One thing I enjoy about Heroku is the option to export code and run a local Rails + Postgres instance if the need comes to fruition. The "escape hatch" is good for any system in which you intend to invest resources, even if you never need to escape. I think AppEngine is a little bit more proprietary, although with some work you could probably move to Python, Django, CouchDB or something similar in an emergency.
At the same time the entry level cost for exploring an idea in code has been reduced yet again. Explore an idea, create something and see what happens. If it works then inject some funds. If it fails, the outlay is time and effort only.