Specifically, the problems with OCSP that Adam Langley talks about here?
Revocation should be done if a certificate is comprised. In this case the valid owner of the certificate will no longer use it. Revocation checks should then make sure, that also the attacker will not be able to use it on intercepted connections.
Langley describes an attack, where the attacker is not only able to redirect the user to the attackers server with the compromised certificate, but also to block or manipulate any other connections from the victim and thus make the OCSP requests for revocation check fail with temporary errors.
OCSP stapling does not help in this case. The server with the compromised certificate is owned by the attacker so the attacker will just not do OCSP stapling, i.e. will not send an OCSP response back within the SSL handshake. This will force the victim to try with the normal OCSP queries and we are back to Langley's attack.
OCSP will only be secure if the client will only continue if it gets a valid response from the server which says that the certificate is ok and if the client can successfully verify the signature of the response. This is the same with and without OCSP stapling. OCSP stapling only speeds up the revocation checks because it saves an extra HTTP request for the OCSP check.