Looks like you’ve discovered one of the many problems lumen and raytracing set out to solve. Without global illumination, something is either in shadow or not. This leads to objects in shadow looking very flat.
The fix before global illumination is to precompute ambient occlusion maps, or use screen space ambient occlusion to darken the nooks.
You’re chasing a golden goose egg if you want a non-static, high performance way to add depth to shadowed areas without raytracing.