It's still a bit scary that they're apparently using some ad-hoc KDF (key derivation function) built from bcrypt and HMAC. bcrypt is a decent KDF, but there are several ways to make a bad KDF from a combination of bcrypt and HMAC. For instance, concatenating bcrypt(salt, pw) and HMAC-SHA512(salt, pw) makes guessing passwords way too fast.

I'm hoping that all they've done is use bcrypt(salt, HMAC-SHA256(salt, pw)) to get around the bcrypt 72 character limit. However, using scrypt would have been better.