How many people should you ideally (when time and money isn't a problem) interview before you could see clear patterns to create a valid persona? I've seen answers between 5 to 30. I'm working on a web site . The client's marketing department have already divided their market into 4 need-based segments so I feel confident to start building personas based on these groups to understand why and how they would use the web site.
I'd say it depends on the complexity of the project. 'Web site' can be easy, but addressing 4 segments doesn't sound too easy. On the user hand it sounds pretty good that the clients' marketing team seems to have a clear picture.
For the number of interviews:
Kim Goodwin has written a useful blog post about it and suggests 12 interviews, but up to x-times 12 interviews, depending on the complexity:
A dozen hour-long interviews are usually sufficient for defining a simple consumer product, though it can take several dozen for a complex enterprise application.
A bit in contrast, a UX Magazine article states that patterns in the data start to emerge after approx. 30 interviews.
My personal opinion is that the most important thing is to talk to potential users/customers at all (ideally equipped with some wireframes), regardless if for the purpose of creating personas or not. It's a good way to get a wider picture and at least some pre-launch validation of assumptions. I am sure you know this but I am not sure if your client does. I would talk to the client and show that there is no evidence or clear direction amongst experts and then recommend whatever you think is appropriate and get their feedback.