It seems to me that sshd should not tip its hand by returning different
responses when a user ID can be used for logins than when it can't --
allowing an attacker to focus password guessing attacks on user IDs with
which it would have a chance of gaining access. For those folks out there
who are more familiar with OpenSSH than I am: How hard would it be to
make the responses indistinguishable?
Are you running the latest version of portable OpenSSH? If not, you need to
upgrade. As far as I know, there should be no more leaks of this sort in
the current code. If there are, please notify the openssh developers (and
include your authentication configuration - your PAM modules may be leaking
the info, and there's nothing OpenSSH can do about that).