While I have no idea how UnrealEngine handle depth I have thought of a technique which would result in that kind of depth output.
The idea is to better manage Z-fighting by rendering multiple (stacked) frustums with shorter distance between ner and far planes. Each step in the cascade would represent one of those frustums. Using near/far plane clipping, stencil masking and optional vf-culling you can easily render the cascade front-to-back, or each frustum to each own framebuffer and consolidate afterwards, with only minor hit in performance.
This should scale better than increasing depth buffer bit depth which has diminishing returns. Besides not all platforms support high bit depth. OpenGL ES require only 16 bits!
It is just a theory though :-)