They have said that they will reserve around 15% of the fuel capacity of a first stage for reusability operations. At the point they need to impart the Delta-V to return to base, they will thus be 85% empty. 

Thus the need for only 3 engines instead of the 9 main engines for the retro propulsion burn. 

I won't try to do the math, but they claim they have, and think it will work.

There was some discussion I did not entirely follow about whether 2 or 3 burns are required to return.  They need to cancel the velocity in the forward direction, and start heading back.  Then need to control entry back into the atmosphere in a survivable fashion. 

Remember they are lofted upwards and forwards when MECO-1 and stage separation occur.  They can ride the upward component and focus on the forward vector to cancel it out.

For Falcon Heavy it is trickier, since the middle core stage burns till much later and is quite a bit higher and faster at it's MECO. 

Some have speculated that launching out of Texas, and recovering in Florida might buy some margin on reusability operations.

For the CRS-5 flight (Dec 2014 scheduled date) Musk has said they will use a barge being built in Louisiana (50mX70m) to try to land on at first, until they can prove to the FAA they can control the landing accurately enough. Musk predicts only 50% chance of success on first attempt. 

By the CRS-8 flight, Apr 8, 2016, they successfully landed a first stage on the ASDS barge "Of Course I Still Love You". They made several attempts in which they literally hit the barge, demonstrating the ability to get back to the barge, and finally the ability to land.  Elon Musk in the press conference after the launch suggested that fully 1/3 of their future flights would use the barge for landing. Usually heavy GEO missions that needed the extra performance.