Well.. I can't really agree with you.
You're basically saying "People are stupid and because of that, it's a bad idea".
The same issues you're mentioning are there for mesh as well.
People may create mesh that's not perfect, they may use their own mesh avatars instead of the standard avatar and people may get confused about it.
As you do with mesh nowadays, you include a note in the product that it's a custom skeleton, include an optional note what bones it has.
There is a very large incentive for using the standard skeleton because most of the products will be using that skeleton. If you're creating a human avatar, there's no point in making your own skeleton for it as there's already one and if you create your own skeleton, you run the risk of people not buying it because of compatibility issues.
The point with a custom skeleton is to be able to make custom style avatars or, in the case where you can rez them, custom style animated objects, like a car for example.
And if you use the same named / IDs of the bones as the standard skeleton, anything that works for the standard skeleton will also automatically work for your skeleton as it's mapped the same way.
I don't think "people are stupid" is a very valid argument.
People will want to create optimized and compatible products in order to be able to sell them to a broader audience. Where this cannot be applied, because their ideas are too outside the box, then a custom skeleton works wonders.
Skeleton A will work just the same as skeleton B. Animating one skeleton isn't different from animating another, mechanically. It just requires support for dynamically changing what skeleton is used for what avatar.
The slider limits will be set by the designer of the skeleton and a created shape will hold the information about the shape for the skeleton it was designed for. Basically it's about dynamic data size as opposed to static data size. Dynamic data size is a little less efficient than static, but it doesn't matter that much for today's processors. Or you can simply limit the amount of bones to a certain number and then have every skeleton be that number of bones if you require static data size.
Bento is an improvement, but needless to say, it has it's limitations. You can't have a finger with 4 bones, should you want that. You can't have 2 heads and 2 necks. You can't put wings on your legs or your head. And you can't make a car (well.. you might be able to, but you'll have 1000 extra bones and you won't necessarily get the exact structure you want)
So while I understand this won't happen tomorrow, I definitely think that this is something to work towards in the future. And frankly, it shouldn't be that hard. Would probably require around the same work as creating a new hardcoded skeleton and support for it, but instead of making everything static, you make it dynamic.