Facebook also has the audience network, which serves ads on other websites and apps.

Then I seem not to understand the benefit they see from tracking non-users' information - how can this be of benefit to advertisers if these users are never targeted? In particular, what benefits would be generated by linking this tracking information with the real identify of a non-FB user (that couldn't also be leveraged in a semi-anonymous way)?

In other words, I see the benefit to FB of being able to say "Person A likes 1, 2, 3; User B likes 1 and 2 - perhaps they also like 3?" where A is a non-user and B is a FB user, but I don't see the benefit derived from identifying Person A to advertisers if they can only target that person on FB, where they aren't a user?

They don't allow advertisers to target non-users. I'm sure they use non-user data for clustering, just as you described.

They also probably use it when the user finally does create a FB account.