pcmattpope said:
There is a physical system that absorbs the energy, and allows us to later recover the energy. It's conserved either as a change in mass, a compressed spring, electrical capacitor, or system we have yet to know about.
The intuition you are expressing here is really local energy conservation, not global energy conservation. This is still true in GR, but there is a key difference between gravity and all the other interactions with respect to it.
The way GR expresses local energy conservation is that the covariant divergence of the stress-energy tensor is zero. This just means that, in any small 4-volume of spacetime, the amount of stress-energy going in exactly equals the amount of stress-energy coming out.
For example, take the case of compressing a spring. We take a small 4-volume of spacetime that contains the event of the spring being compressed; you can think of it as a small 4-dimensional hypercube with faces to the past, the future, and each spatial dimension. On the past face of the hypercube, the spring has a certain amount of stress-energy (its rest mass, since it is unstressed), which is coming into the hypercube (since it's coming through the past face); on the future face of the hypercube, the spring has a larger amount of stress-energy (because it's now compressed, so it has its rest mass plus the stress--note that there are some subtleties here, which I won't go into unless it becomes necessary), which is going out of the hypercube (since it's going through the future face); and on one of the spatial faces of the hypercube, some stress-energy comes in in the form of work done on the spring. So the final tally is:
stress-energy coming in = (rest mass of spring) + (work done) = (rest mass of spring) + (stress in spring) = stress-energy going out
This all works nicely and fits in with the intuition you expressed--as long as gravity is not involved. But when gravity is involved, the tallying works differently from what you would expect from the above (it still works, though, which is an important point).
For example, take the case of slowly raising a ball from the floor to a shelf. Again, we can sum all the incoming and outgoing stress-energy over a 4-volume that encloses the event of raising the ball. On the past face of the 4-volume, we have the ball's stress-energy coming in at a lower height; on the future face, we have the ball's stress-energy coming in at a higher height; and through one of the spatial faces of the 4-volume, we have work coming in.
But now, when making the tally, we have to account for something that didn't come into play in the case of the spring, above: the contribution of a given amount of stress-energy to the tally depends on the gravitational potential. This is because the actual spacetime *volume* of the 4-volume depends on the gravitational potential; more precisely, in order to construct an invariant 4-volume element over which to evaluate the covariant divergence, we have to include the gravitational potential as a factor. So the tally in the case of raising the ball looks like this:
stress-energy coming in = (rest mass of ball) * (potential factor at lower height) + (work done) = (rest mass of ball) * (potential factor at higher height) = stress-energy going out
The potential factor at the higher height is larger, by just the right amount to account for the work done in raising the ball, so the final tally still comes out right: the covariant divergence of the stress-energy tensor is still zero. But the fact that the gravitational potential has to be included to make this work is why people talk about energy being stored "in the gravitational field". Note that the rest mass of the ball does *not* change, and there is no other stored energy in the ball either before or after the process of raising it.