I noticed there are quite a few XYZ compute nodes in the setup and there are also quite a few nodes with internal networks containing displacement functions like cracks. So at first glance it looks very neatly organized and "lean", but I think this is quite a displacement intensive scene all togheter. Anyway...about the "Tex coords from XYZ" nodes...
I do know that computing normals (compute terrain/normal nodes) are expensive, causing the renderer to parse the network upstream 3 times for each normal computation.
This is why the general advice is to perform displacements before compute terrain and preferably just once. Displace -> compute terrain -> displace -> compute terrain etc. will make the number of shader calls skyrocket.
However, I do not know how this behavior is for "Tex coords from XYZ" nodes, which returns the texture coordinates at that renderstage, useful for height restrictions, but not slope restrictions.
Long story short: what happens to your rendertime if you disable all of those?
Yes I know, your render will probably look vastly different and not what you want, but we want to find the cause, right?