Think aloud protocols are meant to render explicit the user thought process as he/she completes tasks.
This is a clear deviation from how things play-out in normal circumstances as users will be attempting to formalise their thoughts in manner that is intelligible for them and for the tester. This will ultimately affect time taken to complete tasks.
In my opinion, A true and viable account of how users complete a task and how much time they spend doing this needs be based exclusively on observation. This allows users to behave normally and allows the tester to focus on what really counts.
Unfortunately I am not aware of any research done in this area but would suggest to use think aloud protocol when you need to conduct quick usability testing.
Also if you do use think aloud protocol try to detect any spontaneous reactions, frustration, surprise etc and isolate these from other user statements because these could lead to substantial changes to the design or IA. Hope this was of help.