As noted, the SpaceX plan is to build a rapidly reusable rocket. That is the only real way to get costs down.
At the moment, no one has ever done what they are trying in several ways. What I find most interesting is the regular testing.
Grasshopper and Grasshopper 2 (officially named F9R-Dev1 and it was tested into a lovely explosion as they hit an edge of the test envelope, the purpose of testing) are testing the control software at the low end of the landing event. (GH completed testing up to 1/2 mile, F9R-Dev1 was lost during testing, and F9R-Dev2 will go much higher in New Mexico).
Every flight they can, SpaceX will be testing the landing system, as a 'bonus' use of the first stage until they succeed. Once they do, they will have the evidence in hand to demonstrate to the FAA that they can safely land a stage on land.
Thus first they will test it over water to demonstrate control and the entire system end to end. Since their original testing, they added a landing barge, the ASDS (Autonomous Spaceport Drone Ship), which provides a 170X300 foot target to try and land on. The first attempt on the CRS-5 mission hit the corner of the barge, but ran out of hydraulic fluid for the grid fins. But boy they got close! Following missions aim to land on the drone ship.
Going forward they are building a West Coast drone ship for Vandenberg launch recoveries.
This gives them additional flexibility. If the ASDS is far downrange and they launch a heavy payload, the fuel required to return to launch site (RTLS) is less, if they actually land on the barge down range.
Testing on the barge makes the FAA certification for landing at KSC much easier, if they have proven repeatadly that they can land on target. (They are building a landing site at KSC at LC-13).
The Falcon Heavy center core will be flying much higher and faster than a normal F9 core, or the two side cores, so using the ASDS for recovery will help payload as well.
Their plans for reuse do involve a 15-30% payload hit but as part of the move from F9 to F9 v1.1 was a big boost in available payload. Thus they will have some flexibility in the future. If your mission requires 100% of the payload, they could launch expendable and charge you more. If it fits in the reuable mode payload range, charge much less.
If they can get Falcon Heavy reusable, then instead of expending a F9 it might be more effective to launch bigger things on the F-Heavy.
The legs on a water landing mission seem sort of silly, but are really needed for several reasons.
1) Test out the leg mechanism on a full up flight, end to end.
2) They need the angular momentum control to handle stage spin that they encountered on flight 6, with the Cassioppe mission.
3) Land on the ASDS and prove their ability to hit a small target to the FAA.
In Sept 2013, they tested the recovery model by firing three engines after MECO to slow down and aim vaguely back to base. Then once it reentered the atmosphere, restarted the center engine (third firing of the mission for that engine!) above the water for the final deceleration, but the vehicle had been spinning and the reaction control system was overwhelmed and could not control it so the fuel centrifuged away from the intakes and the engine flamed out. By extending the legs, like a figure skater, they expect to better be able to control the spin with the existing RCS system and land it on the water.
And if it fails on the Mar 2014 CRS-3 mission, they will try again on the next mission until they succeed. (See above, what no one else seems to be doing, serious experimentation!)