How to make repositories a killer app for scientists

Cameron Neylon wrote a nice post indirectly addressing the question of how Nature might make Connotea more useful. It's well worth reading for its own merits, but I was so taken by his questions, I thought they might be re-purposed to apply to repositories. As Cameron says "These are framed around the idea of reference management but the principles I think are sufficiently general to apply to most web services". The text below is Cameron's, except with chunks taken out (shown by ellipses ...) and some of my text added [in brackets] (added text may replace some of Cameron's text).

1. "Any tool must fit within my existing workflows. Once adopted I may be persuaded to modify or improve my workflow but to be adopted it has to fit to start with. ...
2. Any new tool must clearly outperform all the existing tools that it will replace in the relevant workflows without the requirement for network or social effects. Its got to be absolutely clear on first use that I am going to want to use this instead of ...[some other tool, like a lab web site]. ... And this has to be absolutely clear the first time I use the system, before I have created any local social network and before you have a large enough user base for these to be effective.
3. It must be near 100% reliable with near 100% uptime. ... Addendum - make sure people can easily ... download their stuff in a form that will be useful even if your service disappears. Obviously they’ll never need to but it will make them feel better (and don’t scrimp on this because they will check if it works).
4. Provide at least one (but not too many) really exciting new feature that makes people’s life better. This is related to #2 but is taking it a step further. Beyond just doing what I already do better I need a quick fix of something new and exciting. ...
5. Prepopulate. Build in publically available information before the users arrive. ..."

I think what Cameron's saying is what we all should know by now of the needs of potential repository depositors: make it easier, make it better, make it worth my while.

It's the "how" that's tricky. I don't think this is impossible, but these targets are hard to hit. It's what all the "negative click" Research Repository System attempts have all been about. They do need a re-think, however, if not to fall foul of Cameron's rule #1! So what other ideas are there out there, folks?