There is no security problem with SSH agent forwarding per se; that is, there is no additional risk that you would not see by manually logging in to the remote server and connecting to a third machine.
The risk lies in the fact that a malicious system owner (or attacker, if the system has already been compromised) is in a perfect position for a man-in-the-middle attack.
The MitM attack could come from several different vectors, although two come to mind specifically. The malicious admin could attempt to
sslstrip an outgoing SSH connection so that your data (including credentials) is readable in transit, or he could change DNS so that you are connecting to an SSHd designed specifically to steal passwords, rather than the destination you were initially trying to reach.
Again, though, these risks would be present while using a third-party server even without specifically using SSH agent forwarding.
Hope this helps :)