@dvduval Re: keyword pie
Interesting idea! I am seeing something very similar that started all the way back in March 2010, so it may not be exactly MayDay's feature (else I was volunteered to become an early adopter). Actually, if we go back to that time period there were many reports of this kind of "freezing" of the number of keywords "allocated" to a site.
It basically looks like you may be given a different trust rank for different slices of your "keyword pie". I would not even speculate about exactly how your keywords are sliced - there does not seem to be much logic in which KW goes into which slice. However, as soon as this allocation happened (for most of my sites on March 15th and a couple more sites contracted it a few days later) the "trusted" slice has continued to rank high as if nothing happened. The rest of the pie disappeared off the face of Google.
I have done some analysis of the KWs in the "trusted" slice and the rest of the pie and in my opinion there is NOTHING better about those trusted ones. I have better linked pages with more prominently featured KWs now rank for nothing (went to a bad slice?) and seemingly unremarkable pages (no external BLs, and KWs are sometimes not in the title and only once in the content) continue to rank as usual (stayed in the best slice?).
What is also remarkable is that there appears to be exactly 10 slices in this keyword pie. Only one is trusted, the other 9 are not. You can just take both the number of formerly ranked keywords and total G* referrals, divide it by exactly 10 and arrive at todays numbers. Why 10 and not, say, 12? A better looking figure perhaps?
Mind you, this concept does not eliminate existence of the site's overall trust rank. I also have sites that were sliced first, ranked for the 1/10th of the KWs and then completely disappeared. Most of those were new sites that were probably going to disappear anyway as the "honeymoon" period was getting to an end.
There are also some new developments here. It looks like some sites are now finally getting some of the slices re-evaluated and "released" or "thawed" if you will. This whole concept of slicing may have actually been brought about by the need to limit the load while re-evaluating the entire set of keywords. Even with Caffeine I imagine it would be an unsurmountable task to recalculate trust rank for the entire Net given that most large sites receive traffic on hundreds of thousands of keywords each and there are millions of sites...
Again, great concept but there is still the need to explain why not all sites were chopped up like that. Clearly, not everybody suffered a loss and even I had two sites that were stable during this troubled period. Could high enough initial site trust save you from the "keyword pie" effect?