I explored Google AppEngine for my own amusement a while ago, making a lunch 8ball in the process.
http://blurry-lunch.appspot.com/
The system is very easy to work with, and to hit the ground running.
Limitations I saw in Bigtable mostly revolved around dataset size and access time. Part of the application I was making would randomly choose a location out of a list of locations, to due this I am pulling out the complete list of locations, then selecting a random element in python. As the default index is a non-linear GUID, and I didn't bother to setup a separate attribute in the object, as to find the next available id in the system isn't how DataStore is designed.
The problem is, if you need linear access to a massive amount of records to perform an operation, you could run into a limit where your request takes too long, but what data set would cause such a delay isn't clearly defined, as google's systems are massively clustered.
As far as external Python Libraries, you should be fine, as long as the calls they make are python based. You will have to bundle them yourselves into the directory structure that gets uploaded for deployment.
You should be aware that you will be locking your self to their platform, as there is no production ready system that supports their API. They have a standalone appserver for debug purposes, but it certainly isn't suitable for an actual deployment.
Another thing is that GAE is still in beta, with no commercial support options, and you could not run a wildly successful web application currently without a commercial plan. The limitations are too low to even survive a slashdotting for static content.
EDIT:
Of course this is all as of Feb 19th, 2009. This could all change wildly, or GAE could even be turned off.