What PFS will protect against in this scanario is decrypting SSL sessions whose cyphertexts were captured before the attacker had access to the private key. It doesn't protect against (any) man-in-the-middle attacks.
Before or after they have access to the private key, so long as an active MITM is not performed. In essence, PFS makes it such that no matter what information you have about the server's configuration, passive sniffing of data is not enough to compromise a connection's confidentiality.
Security isn't my field at all, but I'd gotten the impression from HN that PFS is meant to protect against just this scenario. Am I mistaken?