"Filtering just on the interface still leaves open cases where a lot of spam would make the loading of a page extremely slow."
I might be wrong, but don't API calls exist that can pull only particular posts/comments given an author and a permalink?
Given careful coding on the part of the interface developer(s), would it not be possible to make network requests only for comments/posts from users that don't appear on a local blacklist?
Still I suppose you might end up getting thousands of permalinks and account names that have to be loaded. I'd reckon that'd be faster than the naive process of simply dumping the whole comment tree either way.
An API call returning a list of all accounts who commented on a post with single appearances, combined with a call which returns all the replies to a given post by a given author would seem able to prevent any loading issues from spam comments.
These types of calls seem technically possible to me, but whether they actually exist today is a different matter. I have limited experience reading data from public remote RPC nodes, but I'm a far cry from having memorized all the API calls.
These considerations have got me thinking about getting into Hive backend development...
While I'm on the topic, do you have a list of all the API calls supported by your new nodes?