For us, being able to get render elements out of UE4 would be a godsend, and one thing holding back the wider adoption of it. Like has been mentioned, Datasmith-converted materials become too complex and cumbersome to be able to edit in real time, making them semi-redundant, so being able to quickly edit things in PS would be nice. Broader corrections are easy enough using the post-process volume. There is a plugin for camera correction, though I've not tried it.
I don't necessarily agree regarding wages, - what UE4 can do far outstrips a traditional viz workflow, and Datasmith represents a tiny amount of it. If someone is skilled in UE4, they're opening your company up to a pretty interesting variety of work, and if that's something valuable to your company, they should be paid accordingly. If you want to produce real-time VR with interactivity and slick menus, or a car configuration mechanism, or some form of HTML game - or all of these, whilst remaining linked to the viz world, it's hard to look past UE4. Again, it depends on what is valued in your firm, IMO highly-skilled specialists tend to get paid more than generalists. This applies to more than just UE4 though.
For mine, UE4's biggest weakness is in-fact what is perceived as its strength - it can do a lot of different things, but not necessarily that well, as nothing is as simple or well-developed as the traditional archviz workflow (like you hint at being your concern). Even lighting is unpredictable - you need only check out the Epic forums to see people battling to get consistency. Sometimes, things just break randomly and your lighting just stops working. Everything seems to be a case of roll the dice, try things out and see what happens, then make incremental changes. It remains quite a big jump - coming from someone using it full-time for nearly 3 years, I still struggle with it sometimes. For mine, those jobs that need quite a bit of post-pro work or camera matching, UE4 is still difficult to justify implementation. Powerful, but flawed!