That would at most reduce the number of KDF calls an attacker would need to perform by a factor of 2, if (1) the other service is using exactly the same KDF (2) all the usernames are identical in the two services. If not all the usernames are identical, then the factor of 2 is even lower. If an attacker can perform n calls to a KDF then he can also perform 2n calls to it, so if you're relying on that factor of 2 for security you have a problem no matter what. In any case this is easily defeated by adding a random string: KDF(username + "|blablafoo|" + password).
Yes, this would be true in the case of adding a random string as a second salt but then this would have to be sent to the client also.
I am not sufficiently familiar with KDFs but I know that rainbow tables exist to crack straight up md5 hashed passwords for common usernames and passwords where the username has been used as the salt.
Maybe harder to do with a KDF because it's prohibitive to compute that many keys in the first place?
