perhaps an understanding of how a shadow map works could shed some light (or shadow) on things ;)

A shadow map is a precalculated lattace or raster that is used to map a shadow into a scene. it can be more efficient than a ray traced shadow (time wise) but at the expence of memory usage. The importand thing to be aware of with a shadow map is that the ammount of allocated memory is split accross the total cone of the light. So the wider the cone or the smaller the map size (memory allowance) the fewer the squares in the raster (note the word squares).

the renderer checks every element in the raster to see if there is geomitry between the light and a surface. if so it then projects that raster element (square) as a shadow into the scene.

If you can see a square then that meant that the raster is too small and it needs more elements to make a more accurate shadow.

in the attached image you can see on the small sphere how the 256k raster projects a single square into the scene and the shadow doesn't get reasonable definition till its using 2Mb. Conversly with the large sphere you can see the lack of definition of the curved edge even at 2mb (the squares are clearly visible).

I hope that sheds some light on things :)