UE is perfectly capable of high-poly scenes, and using high-poly meshes. It doesn't even have the 60k poly limit that used to be norm for most engines per mesh. Your only limit is capacity of video card memory, just like in gp-gpu rendering. You buy an 6-8gb card,
and you're on good path to create scenes just as complicated as your regular raytraced offline ones.
You can use box-mapping (or any other over-lapping UV technique) perfectly fine for texture mapping, but lighting needs unwrapped mesh because each part of mesh will get 'unique' light baked-in. Obviously the mesh part in shadow can't share UV-island with part that will be hit with light. Therefore the unwrapping needs to be decently strict to avoid issues, coupled with fact the GI baking is done using very crude Final Gathering (Irradiance map)/Photon combo. So not only overlapping, but also seams placement is important.
You can use different UV mapping for your channels. Keep box for Channel 0 used for textures, and unique unwrap for Channel 1 used to light-baking.
That is perfectly fine for all these super-simple apartment demos that are popping up, but really, do this for serious commercial project with set deadline, with tons of unique meshes and small props.
It can be decently automated up to point though, with lot of cave-eats here and there.
The reason is whether you need interaction, and if it can be somehow useful for your clients. Imho, in 95perc. it would be more than counter-productive. Not everything that is fascinating, is also useful and helpful towards every goal.
I love real-time and UE a lot, and did so for past 5 years. But I still chuckle at the naive/kiddish fascination if this is the next big thing that will revolution everything. Of course not.