The whole point of using extern rather than exposing what is actually happening is so that the under-gerwerkkins can change and your code doesn't have to worry about it to pick up the improvements.
That said, CLLocationAccuracy is typedef-ed to double, so I think it's fair to guess that kCLLocationAccuracyNearestTenMeters = 10.0, kCLLocationAccuracyHundredMeters = 100.0, etc. Best is likely either 0, 1 or kCLLocationAccuracyNearestTenMeters, and BestForNavigation is probably one they tossed it to help folks like TomTom, etc.
If you REALLY want to know, you can print out the values -- they're just doubles.
I do not believe that the number of satellites or power to wifi is altered based on your desired accuracy. The way I understand the algorithms, there is an approximation calculation that, the more times through the loop, the more accurate it gets. Hence, less-accurate just bails earlier.
But, again, the more important point is: it doesn't matter. Apple specifically doesn't describe what goes on behind the scenes because that's not part of the design. The design is: if you use kCLLocationAccuracyKilometer, you'll get an answer that's within a kilometer, etc. And Apple is now free to change how they arrive at that without you caring. This sort of isolation is a basic tenet of object oriented programming.
EDIT:
CORRECTION -- I'm just now watching the WWDC session on location (Session 115) and, at about 22:00 or so, he talks about how, when using BestForNavigation, this adds in some gyroscope correction (when available.) However, he warns that, while this is power & CPU intensive, and should be only used when necessary, as with turn-by-turn navigation.
I'm not sure how much more I can talk about this publically but, if you're a registered developer, you can get the sessions from iTunes-U.
(This is WWDC-2010, btw.)