Historical Keyword-Level Data

My firm (Efficient Frontier) manages large paid search campaigns, over $150M in annual spend for about 70 advertisers; I think we do as good a job or better than anyone else out there. We're fortunate right now to have enough business opportunities that we can turn down potential clients when the fit isn't there. In those cases when I've told a prospect that a)we're overkill; b)they have more fundamental issues to address first; or c)we'd rather use our capacity elsewhere, an enjoyable conversation typically ensues wherein I give them all the real-world advice I can.

One of the recurring themes in these conversations is the importance of historical, keyword-level conversion data. Without several months of historical keyword-level conversion data (HKLCD) *no* firm can quickly help an advertiser who's already fairly advanced, *unless* they're in the business of leveraging one client's data for the other (we're not, as doing so would be wrong & quickly lead to most of our clients leaving us IMO).

While stressing the importance of keyword-level tracking months back, it occured to me that there would be sustainable value in an industry-wide HKLCD repository. If you want to jump into search marketing, depending on current keyword generation tools is tantamount to blind dating - not something you want to do unless you're desperate. As an example, Google's Traffic Estimator has been broken since day one; at the individual keyword level it is 30% or more off 70% of the time. Or, think about how illiquid the stock market would be if instead of 8000 equities we had 50,000,000. Now add to that a total lack of publicly-available historical performance data, and you essentially have the Yahoo/Google version of the stock market.

Oddly enough, while aggregating HKLCD data is in the majority of advertisers' interest, it's not happening because:

1)good/early advertisers don't want to help bad/late ones.
2)search engines and SEM's typically are restricted from using client conversion data for any purpose other than helping that specific client. On the SE side I know that's not the case, but it is the case more often than not in the SEM world.

Claria of all people is building an offering of this type, but it doesn't look like it'll be the fine-grained type of HKLCD that advertisers or their SEM's need to launch fully-educated and optimized campaigns at birth. Likewise, the Fireclick Index is fun to look at but you can't actually do anything with it, nor can you trust it (no offense to Fireclick).

Any ideas on how to make an HKLCD repository happen? Does anyone else think it should happen?