Here’s my opinion on the “what to chose?” subject:
Rovers-
Large payload.
Fast.
Stable platform for video.
More payload and larger size allows for more and better sensors.
No frustrating gate generation required.
Hexapods-
Low payload.
Slower.
A bit wobbly for video, though not too bad, since 3 legs are usually on the ground at any one point.
Still a pretty good platform for sensors.
Gate generation takes up a lot of time, but there’s a lot of examples.
IK can greatly speed things up after the initial learning curve, since there’s usually only 3DOF per leg.
Half Bipeds-
Perhaps the cheapest of the servo bots, since 4DOF is all that’s actually needed.
Payload is a joke.
Their speed is arguable, as they’re a new commodity, though I believe that most will be slower than the average hexapod.
One day, we’ll have to put on a race and see. 
Video probably too bouncy.
Not much room for sensors.
Since their mass is usually quite low, the platform isn’t a very good candidate for dynamic balancing feedback.
I say “usually” because there may be some half-bipeds, such as the Scout/209 which are a good candidates.
Since there’s less servos to deal with than any of the other buggers, gate building time is at a minimum.
There’s also a great tutorial for it and a ton of examples (although you’ll find that’s apparent with most of the robots).
Full Bipeds-
Very expensive, since strong digital servos (5645 or better) are practically a necessity.
No payload to speak of.
Snails pace.
Probably too wobbly for any useful video.
Not a lot of room for sensors, and some sensors, like rangefinders and cameras are practically useless.
However, there are a few sensors, such as gyros, accelerometers, and even pressure sensors which can only be fully realised in a design like this that craves balancing feedback.
Gate generation takes up nearly all time not spent on balancing hardware.
Not many examples to be found.
Conclusion:
Rovers are nice and foolproof for learning, getting right to the “good stuff” as well as being the power-house in terms of being able to interact with the real world and accomplish tasks.
Hexapods are designed for those who wish to have most of the above flexibilities without the critics saying that their robot is YARR (Yet Another Rolling Robot).
In terms of learning, it loses the ability to get directly to the “good stuff”, but much of the boring gate-generation part can be accomplished through neat programming and mathematical tricks.
Half-bipeds are a great introduction to servo robotics in terms of expense and ease.
They’re probably the best learning tool to get into this type of thing.
The lack of sensors sort of takes away much of the appeal for some, though.
Full-Bipeds are, let’s face it, only made for bragging rights.
Building the biped makes one an elite, and getting the blasted thing to walk across the floor makes one a god among men.
Dynamic feedback is, of course, the eventual goal of every biped-builder, but it’s one which few progress farther than having good intentions and great ideas.
All-in-all, full-humanoids are an endless moneysink of uncountable frustrations, suited only for self-massochists, like myself.
Still, in the end, getting that bugger across the floorboards is one heck of a rewarding experience.
In the end, it doesn’t matter what robot you chose.
Pick what sounds fun to you and ignore the critics-- they’re just jealous.
Even if you can only manage to turn out with an obstacle-avoiding YARR that burns out it’s motor controller after smashing through every obstacle in it’s path, you’ll still have the satisfaction of knowing that you built a robot.
So, when the little green lizard from that Geico commercial asks you how it feels to be better than everyone else, you can safely shout at the TV that it feels [size=150]FLIPPING AWESOME!!![/size]
, and the rover pretty much is better able to tak eon rough terrain 
]