I like people offering help to others, but I read your article before you updated this... and I immediately wrote you off because of that.
It's great to offer advice, but can't seem to find any indication of what your experience is.
You can also either move the computation to the client over a secure channel, or have another challenge/response method in place. I've actually thought about having a bcrypt challenge/response (from a pool of known values at the server) in order to reduce abuse of an API structure or interface. It was a thought exercise in how to approach such a method better... It was actually a thought on how to approach something like a next generation hive/bittorrent protocol to reduce the attack surface a bit.
In any case, just be careful when you implement more secure hashes with a high compute cost... it can bite you in the ass if you aren't careful.
That's no excuse for using a worse hashing method. You need to rate-limit login attempts anyway to prevent (or at least make impractical) brute forcing / dictionary attack attempts. But it's a can of worms either way (next step is captchas because of tor / anonymizer proxies etc.).
According to http://www.cryptopp.com/benchmarks.html you could check ~6480 bytes of SHA256 in the time that it would take you to just set up your first blowfish check. If you're trying to brute-force an 8-character password, you're looking at 141 cycles per password for SHA256 and 114,950 cycles per password for a single round of blowfish.
GAE won't let you upload and run modules that are C-based, so bcrypt is out of the question.
Don't use a single round of any hash function, including SHA-256.
You shouldn't try to roll your own crypto, and you shouldn't try to roll your own password verification either. Use bcrypt and you'll get salting and you'll get a multi-round hashing process that has the ability to easily scale up in complexity as computing power continues to increase.