Three.js can be as performant as WebGL, since it's not hard to drop down to buffers and shaders and sidestep any per-frame overheads.
There are tons of gotchas though, especially for VR render loops. If you naively add even a medium amount of stuff to the scene graph, it's easy for the management overhead to eat up your budget.
On the plus side, it's all Javascript, the code is solid, and it's easy to profile with browser tooling. But Three.js isn't going to do any voodoo magic for you if you don't know what you're doing, whereas Unity might. A double edged sword.
I have a few demos with physics running at an easy 90FPS on an HTC Vive. The same, exact demo ran at a rough 75FPS on the Gear VR, but that was a while ago and I understand Oculus has released some performance improvements, so it might be a solid 75 now. The difficulty is mostly in keeping the number of draw calls low (which is common to all VR apps), and avoiding any buffer copies (which is unique to WebGL). I've been trying to build some best-practice oriented components into my environment because of this.
I don't wish I hadn't chosen the browser because then I wouldn't be in the browser.
There is a lot of good that comes from being in the browser and using web technologies. Progressive Web Apps, when done right, are a great user experience compared to downloading apps out of an app store. Web apps are far more discoverable than apps in the app stores, as your web app is constantly being crawled and indexed by a plethora of competing search engines. You can share a web app with someone by just sending them the link. You don't pay a hefty vig to any app stores just for the pleasure of getting buried at the bottom of the new releases list after a few hours. You don't get throttled on how often or delayed on when you can update. You don't get censored on content. It's completely cross-platform. Everyone is allowed on the Web whenever they want.
I do a lot of work in Unity, too, and I like some things about it. I like working in C# over JavaScript. It's certainly somewhat easier to get good looking graphics, but definitely not a lot easier. You have to have good art assets to begin with, and the whole interaction has to be designed around how Unity's asset pipeline works, which basically means you're locked into Unity harder than most any other proprietary platform I've ever dealt with. It's a lot worse than, say, trying to migrate from Oracle to Postgres, or something. A lot of times, a few, small compromises on the art can get you 90% of the graphical quality on WebGL. Most of the really awesome Unity demos you've seen are as a result of their material and lighting baking process and are actually very static. Unity's primary value is its asset pipeline. The graphics could be done just the same by just about anything if they had a decent asset pipeline.
And it's not even really all that great of an asset pipeline, it's just a lot easier than hand-baking all those textures. But it comes at a cost. Unity doesn't play well with source control. Collaborating with others is fraught with peril. It doesn't deploy over the Web very well, either. There's no automated testing in Unity. There's no continuous integration. You can pay extra for a cross-compilation service, but unless you pay big bucks you get queued for quite a while. If you don't like something in Unity, you're kind of stuck with it. With WebGL, if I don't like Cannon.js for my physics system, I can replace it with something else.
So for my WebGL projects, I actually don't care about graphics that much. I think the focus on graphics is actually a significant harm to VR right now. I'd much rather have flat-poly cartoon graphics that were cheap to build and leave me at least a few MS of processing time per frame than so-called "AAA" graphics, if it means I don't have any app cycles left or I start missing frames and really need to lean on Asynchronous Space Warp to fix it for me.
Take the course that Unity offers when you purchase a pro subscription. There are a lot of super arbitrary and half-hidden details that you just want to go directly to the source to find out about them. The hardest part is staying awake for it all. The narrators voice is extremely boring and slow. I personally had to set the playback speed to 1.5x just to not fall asleep during the videos.
