Lavabit was designed to protect the privacy of e-mail by allowing users to encrypt messages stored on the Lavabit servers. Once encrypted, an e-mail could only be decrypted with a user’s password. The system was made to protect messages on Lavabit’s servers from prying eyes. Quite simply, the goal was to remove Lavabit from the surveillance equation.
Quite simply, Levison failed to remove Lavabit from the equation, because Levison was only willing to provide privacy to a point where it didn't compromise convenience.
Fun fact: Lavabit turned over an archive of encrypted emails from Snowdens account when he was first contacted by the feds. Those emails are still perfectly secure, because Snowden is the only person who knows the password required to decrypt the mail files. The reason the feds demanded the SSL keys (a few weeks after he first turned over the emails) is because they could not decrypt the mail files and wanted to try and capture Snowdens password the next time he accessed his account.
Erm, that's only the case for systems that use cleartext in transit. If you encrypt and decrypt (and we assume you're doing it right) at the endpoints then the intermediate can't do anything with the content.
Your 'no matter how' bit is correct, and it was flawed to rely on SSL for transit only and hope the law would protect the keys. Hence tqbf's point that using cleartext in-out is just plain bad to begin with.
This is the part that baffles me. The lavabit about page clearly identified the FBI and NSL as a threat. Lavabit was supposedly designed to be secure against the FBI, even if the FBI didn't play by the rules. But then he still expected them to play by the rules?