I've had the (still fairly unique) opportunity of actually having to make very different considerations with regards to user experience. This came in the form of designing for HoloLens. I bring this up because I believe in 5 years we'll be much closer to realizing the potential of VR and MR (mixed reality, like HoloLens.)
I do think VR devices will live a short life and die off. VR (in the form of a single purpose device) doesn't really have a place once a device can provide both augmentation and a totally virtual experience. I don't think we're far away from that.
Certainly what we'll see is a uptake in MR devices. I suspect future iterations of HoloLens and other MR devices will bring forth a desire to experience and incorporate MR. Some businesses are already jumping on the opportunity, but in my opinion the technology isn't quite there. In the case of HoloLens, it's clear it still lacks a real understanding of intuitive input.
I think one of the big mistakes people make when considering new UX patterns in 3D space is that not everything is designed for 3D space. And conversely, not everything designed for 2D interfaces works in 3D. You can't just move your app to VR. Certainly you'd think it would be rather strange if a restaurant provided its menu as a stack of blocks. Some interfaces are indeed better suited for 2D, so for that reason I believe there is a place for some 2D interfaces in a mixed reality future.
The other big mistake I've found people make is the idea that in order for something to have great UX, it should mimic real life. Perhaps this is true in some cases, and if you're building a product that's designed to mimic real life then that's probably the best choice. But new experiences will undoubtedly emerge (more often than not, I suspect), experiences which are foreign in concept to us now because its simply not possible in our physical world, and that will be a real test for UX experts out there.
One thing we will need to focus on is understanding what it means for an interaction to be discoverable. A lot of people seem to think voice is intuitive, but I don't think that could be further from the truth unless you've got a general intelligence to talk to. No one thinks automated answering systems are user friendly. Even with general intelligence it can be hard to put your intention into words when it comes time to "take an action."
Personally I think the optimal solution for these types of interfaces will be a mixture of context awareness and neural impulses. If I can look at a TV, and the device can see what I'm looking at (on board camera also sees a TV) then it has an understanding of what actions I might be interested in performing. At that point it can show options in 2D above the TV or however you want to lay it out. I'd then be able to look at the option I want (device tracks pupils and knows with accuracy what I'm looking at) and think about touching it. This impulse acts as an "invoke" action on the current thing I'm focused on.
If this stuff becomes possible, then that's about as low friction as I can think of without interfacing with the brain. Will be be there in 5 years? Hard to say. I'd be willing to bet we might have something that gets us partly or mostly there, and may have to be tethered to a secondary device like a phone for additional processing.