That PBKDF2 has been more extensively studied, but it has lower memory requirements. So the chances of a surprising vulnerability being found in bcrypt are higher, but failing that, it is the better option.
In what way has PBKDF2 been more extensively studied? It's the product of a standards effort, not academic research. bcrypt was introduced in a refereed Usenix paper. I'm not suggesting bcrypt is particularly rigorous either, but I'm unclear on what "study" you're referring to for PBKDF2.
I'm also unclear on which "assumptions" you believed bcrypt depended. Could you be a little more specific, please?
What's maybe more relevant is that PBKDF2 is explicitly called out by name as the Proper Thing in various standards, so it's less work to use it in environments where people care about those standards.
