Heh, perhaps I'm just overly enamored with Delicious.
Delicious had different goals than we do. :) I think a good way to look at this is to ask yourself how tags are currently limiting or otherwise affecting your experience, and what specifically you'd like to change or accomplish based on that. There are endless things that we can do. But the key question is why we do what we choose to do.
Oh, but we can make similar claims about anything we don't like :) Sometimes I look at a change and want to say, "but is this the most pressing problem we need to work on?" But at other times I like a change and I say, "I don't know what problem this solves, but let's put it out there and see how people use it." Both are forms of rhetoric cloaked in the semblance of objectivity, thanks to a biased default. Discoverability is a problem we'll have more and more as the community grows. Tags are a way to do discoverability. But they add a concept to the user's cognitive model. Add enough concepts (hide, ignore, mute, ..., yuck) and the whole thing starts to break down. Delicious showed a great way to have a parsimonious set of nouns and verbs that can all be applied to each other. I don't know what goal it set out to accomplish, but I've seen different people use it in very different ways. So it merits our attention.
I hear you. But in all honesty, I've found that the changes that we've made that end up being most satisfying were those where we could clearly define to ourselves the need. It's a lesson I've learned a few times now. Columns and and community tags are the first examples that came to mind where a large aspect of its nature was just satisfying curiosity. Experimentation is not all bad, of course; sometimes it opens our eyes to what we'd really like, but it can be a distraction too. I agree that discoverability will likely become more of an issue as the community grows, and I also agree that it is best if we can limit the cognitive load required. Above you mentioned to b_b that tags are slightly unsatisfying as they have an asymmetry with the first-class user. b_b makes a good point that they aren't sentient, or they don't create content. There might be something fruitful found in the tags that users follow and use. We have relationships between users, and we have tags that they follow and use. Maybe tags can have some sort of autonomous nature to them? Maybe posts earn tags, based on user activity? I'm not sure. I'm just turning it over and seeing if it might fit into what we have in some surprising way. It so easy to expand upon tags, and I see that as not only increasing the cognitive load, but pushing their importance up to a point where it might diminish user relationships. I'd most prefer we hit upon something where we all go: oh yeah! and it just fits like a glove.