It seems from these (preliminary) results that there is alot of interest in the soft shadows. I don't know how much research you have done regarding shadowmapping, but from my own experiments it turns out to be non-trivial. Not sure if you need it, but here is some advice:

For the last week or so I have been working on a shadow mapping shader. They are not quite soft shadows yet, but as it turns out that is the least of the problems. When using the shader on any scene larger than the one you can see in that thread, serious aliassing problems appear, even when using a huge shadowmap. This is not just a problem of my shader, but a general problem of shadow maps.

Are you planning to create the shadowmapping shader for sunlight, or just small spotlights/omnidirectional lights? Spotlights are not so difficult when they are used on a small area and distance. The same goes for omni lights, but here you will have to use a cubemap or a Parabloid map. Sunlight/directional lights are the most difficult.

Problems arise when a single shadowmap pixel is projected on multiple screen pixels. This happens when either the surface is (nearly) parallel to the light direction(so there is little space for that surface on the shadowmap), or when the shadowmap resolution is simply too low to cover the area. Another problem is that of 'surface acne': false self (un)shadowing. This happens because of limited numerical precision in the shadowmap or the calculations.

To solve these problems, there are basicly two solutions:
-warping: using a different projection matrix to make optimal use of the shadowmap space.
-partitioning: partition the scene/frustum in some way and using a seperate shadowmap for each one.

Here are some papers on warping and partitioning techniques that look promising:
Parallel-split shadow maps I'm currently implementing this, I can send you my code (although it's quite messy at the moment..).
Perspective shadowmaps First warping technique. Seems to work OK but also difficult to implement. Nevertheless, it's usefull to read this to understand the other warping papers better.
Trapezoidal shadow maps Looks like the best warping technique atm, but it's patented (just like the z-fail shadows implemented in the beta, btw!)
Light space persective shadow maps This is another goodlooking warping technique that is not patented.

There are tons of other papers adressing these problems, but many cannot be implemented in hardware/realtime rendering or simply don't seem to yield good results. Hope this was of some use..