I have lots of opinions but I will save most of them for my post on the "streems" standard.

My main criticism is that I don't like the idea of forcing this data to be on the blockchain. I'm okay with optionally putting account following data on the blockchain, not necessarily for generating feeds for yourself in the client, but as a signal to let other people know who are the interesting people they think are worth following. But I think the feed definitions should ultimately be private.

Well, you could still share the feed definition if you want. In fact, maybe the feed definition could even be put into a custom JSON operation to easily share it with other people with a simple link. But it shouldn't be necessary to put it on the blockchain for steemit to provide the feed of posts/comments to the end user.

Steemit already can generate a feed of posts/comments with any of the supported sort orders for any chosen tag. Steemit could easily support doing the same except for any chosen user. The client could then merge all of these feeds into one, while filtering out redundant posts/comments as well as any satisfying some predicate on the metadata of the post/comment (author, tags, net_rshares, depth, last_update). To save network bandwidth, the actual content would not be provided in all the individual feeds (just the metadata including permlink). The client could then concurrently request from steemit.com servers the content for each post/comment that makes it into the final merged/filtered feed (and perhaps even grab it from cache if available and not invalidated).