A lot of real phenomena can be simulated in 3d now, but breaking things is one area that hasn't  had a lot of development. The only software out there (that I know of) is Blastcode, but I think it still requires a lot of user control, and it's only available for Maya.
Has anyone else thought about this? For example, if you wanted to simulate a hammer striking a concrete floor and shattering part of it, and not having to animate much besides the hammer's initial momentum:
The simulation determines the hammer's momentum, mass and surface area, along with the point of impact on the surface of the block of concrete (obviously your floor has to be a 3-dimensional object in the scene, not just a plane.)
Pathways of stress fractures could branch out from this point, determining the new surface geometry. Maybe? haha.
There are other things to take into account, such as compression, pulverization, the fact that many solid materials such as concrete are not uniform in density, and a lot of other stuff I'm sure I'm overlooking.
I don't know how UV's would work with this. Or if you could simulate with particles that behave like solids.
The other big thing, I'm guessing, is that this obviously deals with solids, and many of our programs only understand geometry in terms of surfaces. Maybe that alone is the breaking point.
The only R&D on this that I've seen so far is on Ron Fedkiw's page: http://physbam.stanford.edu/~fedkiw/
What does anyone think? Is this possible, or at least a few years off?

I'm kind of surprised that this isn't more commonly done already, given that fluid simulation is now readily available. Maybe that just proves how little I understand about the subject. :) I imagine that some sort of particle system would be a good starting point, to overcome the initial hurdle of not having a solid object to work with. Ron Fedkiw is using tetrahedral (ie quasi-solid) meshes in at least some of his research. Then it would need a physics simulation which made the particles stick together in the right way until struck. My guess is that you won't have to wait as long as "a few" years.

That would be great, I hope you're right.

Here are two of his papers dealing with the methods involved:
Fracturing Rigid Materials:
http://physbam.stanford.edu/~fedkiw/papers/stanford2006-07.pdf
A Virtual Node Algorithm for Changing Mesh Topology During Simulation:
http://physbam.stanford.edu/~fedkiw/papers/stanford2004-01.pdf

Software to do this sort of thing does exist (see -> http://www.rockfield.co.uk/applications.htm) but it's double plus uncheap ;)

I haven't come across any free or OSS programs that offer the facilities to do this kind of simulation.

No matter how much or how little you pay, these simulations need a fair amount of time to setup and a lot more to run.

I wonder how much CPU & RAM that would eat. Think it's slow enough to render something as it is with some programs, imagine having to numbercrunch fractal based crack propagation and surface generation...

Even if you could do it at home, you might need to build a farm to make it workable.

Quote - I wonder how much CPU & RAM that would eat.

Trust me, it's capable of using a lot.
Run times of several days wouldn't be unusual.