Personally, I think that a beta release is the only way to fully test an application.  No matter what your budget and resources for QA may be, I don't think it's realistic to even attempt to test every possible permutation of machine configuration, interaction with third-party apps, environment settings, etc.  And some applications have functional requirements which are intrinsically difficult to replicate.  Can a test platform accurately reproduce an environment of a couple hundred thousand users simultaneously accessing your service?  Are your 100 terabytes of randomly generated data going to cause the same reaction from the servers as 100TB of individual entries? Maybe.  But I think that the increasing complexity of the environment in which applications operate increases the tendency towards chaotic behaviour.