Some feedback on the current SRP design: It would be awesome to have a bit more rendering and culling control. We have a scene with quite a bit of foliage which is sort of streamed into the scene. Naturally we have opted for a solution where we use Graphics.DrawMeshInstanced, since using gameobjects would kill performance (quickly). To be able to support culling and LOD, I feel like I have basically re-written a lot of functionality that already exists behind the scenes in Unity as well as creating my own rendering thread.
It seems like it would be fairly straightforward to add an API of sorts that allows injecting custom rendering into a scriptable rendering pipeline where we can leverage the existing culling/LOD infrastructure that must clearly already exist native land of Unity? I remember seeing a thread about persistent drawcalls before SRP was born. That could be one way of going about it? Submitting a list of renderables (mesh/materials) that then gets culled/drawn?
It would also be nice to have some more control over UI rendering (like I have mentioned here: Feedback Wanted: Lightweight Render Pipeline page-4#post-3924151).
(Also BatchRenderer.Flush seems a bit on the slow side, right now?)
Apart from that, our studio is looking forward to releasing our VR game that uses the new SRP system, early 2019 (we keep our fingers crossed that we get a non-preview version before launch). It has really enabled us to push performance and pull off some neat rendering tricks, compared to our earlier releases.