Surface significant pending features as a first-class entity within the Stack Exchange network with mechanisms for community input and collaborative decision making. One form this could take is a "uservoice-like" site where the items are those that have gained traction in meta or are proposed by the team (i.e., user suggestions aren't directly allowed), with related chat rooms and/or meta questions and up/down voting.
Let me describe a scenario that I think happens too frequently on Stack Overflow. A feature appears, seemingly out of nowhere; lots of people flock to meta to complain about the change; people who frequent meta a lot say "what are you whining about, it was discussed in question blah, blah, blah"; you navigate to question "blah, blah, blah" and you find a feature request to which @Jeff has responded (as often as not, not having the most upvotes) and, therein, you see the origin of the feature that just appeared.
From the outside two things appear clear. First, users are genuinely surprised by new features and changed behaviors. Second, the team is genuinely surprised that people (who care) aren't aware of these things. I attribute this to the difference between the way normal users use meta and the way the team does. Normal users don't spend a lot of time in meta; maybe we come here each day and browse around, but we're not driven to dive deep into the question pool and participate - if you're not on the team and you do this, you're left to draw your own conclusions about your normality :). For team members (and mods), meta is much more important and they do use both tools and searches to dig into meta issues and to expose ideas to the meta audience. The audience those requests finds probably depends on the time of day and, likely, the amount of attention those users give to meta (see previous aside on normality).
Additionally, feature requests are mixed in with, on meta, a vast array of other types of posts, including complaints about previous requests, support issues, bug reports, requests for features that will never see the light of day, duplicate requests for the same feature, etc. Unless you take the time to highlight or search for feature requests AND wade through the resulting morass of responses, it's easy to miss a feature request. If you don't log on to meta regularly, you have no chance - for one thing, features sometimes seem to appear within hours or a couple of days of the request being made.
My suggestion is to elevate feature requests into their own site, not as a regular Stack Exchange site, but one with a different mode of operation. Once upon a time we had UserVoice and, while it worked pretty well for certain types of questions, it failed at being an all-around support site for SO. It suffered from some the same problems I describe above. With a few tweaks, though, such a site or something similar could be made to work better as a way to collaboratively give input on the direction of the Stack Exchange platform. The key features that would be necessary are:
- An intake mechanism from a larger, freeform suggestion forum (meta)
- A winnowing process, perhaps within the team, from proposed to tentatively accepted features
- A related space to take discussion out of the way, i.e., we want it to be easy to see, navigate, and vote on the collection of tentative features
- Voting (unlimited? limited by rep?) on tentative features. On UserVoice we had 10 chits we could hand out. I think people ought to be able to vote up or down on each item, but I'd be open to ideas that tied the number of votes to participation.
- A means of exposing which features have been accepted. Knowing the roadmap for implementation would be nice as well.
Additionally, there would be no ability to add items (except for the team), no editing of existing items (except for the team), no answers (just voting), and no inline comment streams (just a ballot with links to offline comments).
Exposing feature requests in this way would give people who are interested (passionate) about the direction of Stack Exchange a more convenient way of participating in its direction. By letting users continue to propose features on meta, we don't place any artificial barriers to input. By only putting the realistic feature requests in the tool, we make it easy to find and give input to features that might actually have a life. Using voting (up and down) we can both provide guidance into the relative importance of said features and whether the features actually make sense, perhaps leading to more refinement. Moving discussion completely out of the way (into chat or a related meta question), we keep the browsing and voting interface clean.