The goal is a rapidly reusable launch system that can be deployed anywhere in the solar system. A parachute system would work well here on Earth, but the martian air is too thin for it to work there. Therefore they are aiming to make it work straight off of rockets.
As for the fuel issue, if I remember Elon's figures right, SpaceX currently takes a $100 million rocket, fills it with $100k of fuel, launches it, and throws away the rocket. Hence the desire to reuse the rocket.
He's far from the first to try this. But the potential failure mode that others encountered is along the lines of what you indicate. You wind up with a reusable rocket that gets to altitude and back, but which is unable to lift a useful payload on top of that. He thinks that he can avoid this, but has not proven it yet. However even if he needs 10x as much fuel per pound lifted into orbit, he's still saving money. (His actual goal is, I think, 10% of his current launch cost.)
IIRC they are expecting a 10% drop in total cargo to LEO when the F9 is used in the reusable configuration. It takes far less fuel to get down than it does to get up since they only have to decelerate the lower stage (not the upper, or the cargo) and since the lower stage will be mostly empty at that point (much lighter). Furthermore the initial decent burn will be at a much lower outside atmospheric pressure (near vacuum?) resulting in a higher efficiency.
I don't know how SpaceX is planning on handling that. Maybe the trajectory they put themselves on is going to be slow enough to survive reentry using just the main engine nozzles as make-shift heat shields?
Maybe they are going to push the gravity turn later into the flight so that the first stage doesn't participate in that as much?
This should allow mass to orbit at 1-10% of current cost.
The big problem to be solve here is firing a rocket while traveling backward at high speed. If they can pull off the Phase 3 test (high speed descent burn), space access is revolutionized (a space business boom):
* "transhab" space hotels (closer than you think, Bigelow already has two test stations in orbit), * lower cost human trips to mars, * massive increase in number, size, and speed of exploratory probes, * communications satellites for everything (Google in the sky).
http://en.wikipedia.org/wiki/Grasshopper_(rocket)#Test_phase...
Not true considering you need to bring that extra fuel up in the first place, so you need a lot more fuel just to carry the weight of the descent fuel into space. It is a really tricky cost/benefit equation, but Musk seems to have figured that out.
Once you are able to re-use the hardware then the cost to orbit can be lowered, which means that the 10% payload cut you took earlier actually isn't all that expensive in the first place.
A Falcon 9 lower stage also isn't a spaceplane covered in very expensive and fragile tiles, sitting next to a rocket. That helps.
The SpaceX project suffers from none of these problems. A relatively conventional rocket with medium-performance engines, no crazy DoD requirements, and a sane budgeting process should do a whole lot better when it comes down to being actually reusable.
Though it may go into "space", a Falcon 9 first stage will not go to orbit; it is decidedly suborbital. This makes it much easier to recover, as it is not going that fast (for a rocket) at engine cut-off. (The second stage takes the payload the other 2/3 of the way to orbit. While they imagine recovering the second stage, that's a lot harder.)
Being suborbital, the first stage probably will not do a retro-burn, as it will come back to Earth naturally. Entering the thicker part of the atmosphere it will reach its terminal velocity, possibly falling on its side to make that slower. They might use drogue chutes for this, or might not.
Once it has fallen to some frighteningly small distance from the ground, it will right itself and use a small fraction of its launched fuel (probably close to what current F9 stages keep in reserve) to slow down to landing velocity from terminal, steering along the way to land exactly where it needs to go.
I once did some rough numbers to guess that the landing propulsion will take 10-20s with 1-2 of the motors, using 5-10% of the fuel. (With only 5% fuel, a rocket stage is actually crazy light, which makes this work.) Hair-raising, but well within the capabilities of today's sensors, actuators, and computers.
No one's actually done this yet, but it seems plausible. I worry mostly about the stage surviving the initial re-entry, because reports are that earlier F9 first stages, which were supposed to water-land on parachutes, were totally trashed, possibly during that event. SpaceX actually has that data, though, so that's either not the case or they have some plan to make it work.
With that, a question of descent, once it's reached orbit and released its payload, would it re-enter with a short burn towards its earthly destination, decelerate with, say, parachutes, and then do its final descent onto the pad with the rocket? I can't see it doing its entire descent with a rocket, that would use so much fuel, and require so much extra weight.