There is quite a bit to unpack here so let me try to break it up:
I have built in Second Life. It is easier than, say, Unity or Unreal, which I have also used. I found Sumerian easier than all for specifically building singular interactive online experiences. It is not a world so it is a bit of apples to oranges to compare it to SL but it can do things like present the user with a virtual concierge that can hear and respond in multiple languages, pull data, have animation and gestures - all for a new user. it is an impressive and fairly easy tool.
I'm not sure what you get this from. Users will always come up with their own use cases. If your use case is ostriches in lingerie, then have at it. And if you say that that's preposterous then you obviously haven't spent enough time in SL because holy crap some of the stuff in there is weird. But if that use case has a very specific interest for the user (and specially in Second Life, the interests can be very, very specific) then it would lack the general appeal necessary for my stated goal, which is mainstream appeal. That doesn't make it wrong, just wrong under my parameters.
That's what you think will work and take off then advocate for that =) And if you are right and my mainstream utilitarian approach is wrong, then mea culpa. If the end result is VR/VW use by the masses, then I will be glad to have been wrong.
I don't think I was ever that specific. I think business uses will legitimize the platform with a lot of folks, as well as give users a reasons to come into the world and get familiar with the UI, get comfortable. But of course I see entertainment and social uses as well.
It would be hard to show research results for a build case that hasn't been tried yet. There are several groups trying the user generated approach and they seem to keep very modest numbers amongst a very niche crowd, according to the trends and data. I am talking about AltspaceVR, Decentraland, Sinespace, Mozilla Hubs, etc, which are some of the major players I keep tabs on as far as updates and repeated user numbers. That's why I propose a different method.