I think one of the mistakes may be that they focus on a wrong metric. I alway read signups/users as a cited metric. It would be way wiser to focus on improving more active metrics (engagement stuff).
The start page is a good example...sure right now it seems to optimize for account creation but I'd say it would be more of a win if users actually engaged with the twitter stream. If it's useful they'll sign up.
[alas I hope all of this was discussed and tested to death already]
I assume that metrics are why every site I go to has a glossy landing page with a huge "Sign Up" button, and then the "sign in" link is hidden in the corner. I'm sure that passes all the A/B tests but it gives me the impression that returning Web users just aren't that important.
A site I signed up for recently at least lets you log in by typing your existing username and password into the huge sign-up form. That was a breath of fresh air.
[alas I hope all of this was discussed and tested to death already]