It seems to me there are a couple of "use cases" we need to tease apart.

If the goal is to digitise and aggregate (e.g., transcribe museum labels, aggregate from old museum databases) then ultimately yes, you take whatever people give you and clean it. Ideally you'd give people tools to clean stuff up before they serve it, but you try and make the barriers as low as possible. 

But once you have that data for the love of God (oops, rant) don't simply repackage dirty data. What  motivated my initial tweet was having to deal with a mess from the EOL API where names may or may not have authorities, depending in part upon what the source of the classification was. It's a mess, and I don't understand why people insist on serving this stuff.

So I guess my point is that if you're a primary provider (i,.e.,  you're the person or institution that digitised the original metadata) then you have an excuse, but if you are serving data you got from someplace else, give me canonical names.

Regards

Rod