1. Google App Engine (GAE) is non-relational in both structure and use. If your app has significant database complexity then your GAE database may not be consistent at any particular time. Correcting those inconsistencies is up to you.
2. Each year computers get faster and faster. What doesn't scale today may likely scale tomorrow. Sometimes waiting six months and buying a faster server is a better solution.
3. The a priori likelihood that your site will be popular enough to demand the scalability GAE can apparently deliver is very low, indeed nil. Are you willing to pioneer on an inconsistent infrastructure when you could use a proven relational one?
4. Above I said "_apparently_ deliver...": it is by no means a given that GAE will prove adequate. It is, after all, a beta site.
5. Is it worthwhile investing the time to learn a new non-relational database system and framework and implement your site in that system when you _know_ you could implement it instead with proven technology? Were your site to prove popular you might have to buy some computing power, get new servers, etc. But isn't that a known path? Isn't GAE the unproven, unknown path? What do you do if your GAE app doesn't scale? Is the flaw in your code or in Google's infrastructure? You can't always tell because Google's infrastructure isn't transparent.
I see GAE as useful mostly for it's integration and toolkit. I would welcome the addition to GAE of a relational database (or of even a non-relational database that guaranteed the ACID (Atomicity, Consistency, Isolation, Durability) properties. But barring such an addition, life will continue just as before GAE.