We still have a big issue with perf with shadow mapping.

Calling screenspace reflection an advancement, hmmm.. at least it is a workable hack. Nevertheless, shadowmapping performces really good nowadays, but it scales with scene complexity and number of shadow caster. Scene complexity for shadow mapping (no pixel shader needed) can be handles by even older GPU really good, the problem is, that the number of potential shadow casters really increased the last years (10 light sources in a scene vs 1000 in a modern game).

With the introduction of deferred rendering , the scene complexity got less important over night, enabling suddenly large amounts of light sources with the result, that a lot of other effects got broken (eg reflection got really expensive).

The next big thing could be voxelation of the scene and raytracing. This would enable screenspace raytracing to improve the many light dependent effects, like global illumination (bouncing light), shadows, reflection. Eg an early version of UDK4 got some advanced voxel/raytracing stuff in it which looked really cool, but the leave it out of the final version. The issue is, that a voxel representation really costs a lot of memory and/or GPU bandwidth. There's research done in reducing the memory impact (sparse voxel trees etc.) and to improve the raytracing performance (cone raytracing). I hope that we will see some of the more advanced stuff in the next years.

I agree with Ashaman that voxelization could potentially allow for new ways of approaching shadows, at least if games have that data available to them. The recent trend of calculating occlusion from analytical occluders (like what they're doing in The Last of Us) is also an interesting development, and potentially opens the door to completely different ways of approaching shadows.