Very interesting subject. I was looking for the updated versions of that study, thanks for sharing!
About the odds, I think there were some numbers provided by Kaufmann in the Rybka forums, at the time Rybka was playing GMs with pawn, pawn&move and the exchange odds. 500 points look pretty decent to me, but this probably should get higher as the rating of the stronger side increases. I doubt Kramnik would survive against a 2285 rated player with a piece down.
I haven't been to the Rybka forums in about year -- I guess I forgot about it...
I'll have to go back there and check it out.
And I'd agree about Kramnik loosing to a "mere master" at knight odds. Although I couldn't say I'd be surprised if he managed to win anyway 
I suspect the 500-point difference between Kramnik and a 2285 player is more in practice than the difference between players rated 1100 and 600. The lower the ratings, the more blunders occur, so the difference of a Knight or any odds will mean less to the outcome.
Because chess ratings are merely a statistical formula, the chances of winning between a 600 player and an 1100 player are exactally the same mathematically as between a 2300 and 2800. But like you said in practice it is different (due to the k factor and sub 1000 rated players being underrated in general).
And yes, I agree that depending on the rating, material odds will be very different. Surely even piece odds at the 600 level means very little while at the GM level the difference is tremendous.
I figured it would be possible to work out some kind of material to rating relation by comparing the average rate of error (according to a program) over the course of a standard game between differently rated players; but
I'm too lazy / not birght enough to do it myself, and figuring someone else has already done the footwork, started asking around.
[DOUBLE POST] 
Because chess ratings are merely a statistical formula, the chances of winning between a 600 player and an 1100 player are exactally the same mathematically as between a 2300 and 2800.
I don't agree with that. The higher up you go, the more each individual point is worth...so the difference between 2300-2800 is larger than between 600-1100.
They modify how quickly you gain and loose rating points, but it's still a statistics formula. In practice I'm sure a 600 player can upset a 1100 player much more often than some master can upset Kasparov (probably never). Also the amount of understanding between a 600-1100 is nearly none, simply pay better attention, while the amount between 2300-2800 is monumental.
One slight error in this discussion is that the rating difference does not estimate the chance of winning, rather it estimates the expected score. It's a subtle distinction. Indeed, if a win and a loss were the only possibilities, they would be the same. But in Chess things are complicated by the possibility of a draw.
So the expected score for a 600 vs. an 1100 is the same as the expected score for a 2300 vs. a 2800. That's the way it works. If that wasn't true, they wouldn't have those ratings. Say you had those two pairs of players play 10 game matches, you would expect the higher rated player in each case to have a score of about 9. But that doesn't mean the individual game results are going to be similar. The 600 will probably beat the 1100 in one game. However the 2300 probably won't beat the 2800, it's more likely that he will draw two games and lose eight.
And of course the character of those games will be totally different. The lower rated games will be determined by who drops the most pieces while there will be no pieces dropped at all in the higher rated games. But the ratings say nothing about the character of the games, only the results.
Consider an 800 rated scholastic player who drops a knight for no compensation against a similarly rated opponent, how much will being a knight down effect his odds of winning? Clearly when a master makes such a blunder, resignation is justified, but when a beginner does this it's not such a big deal.
Ratings are simply statisitcs. They try to predict the outcome based on your past preformances. A 100 points difference gives the higher rated opponent a 64% chance. 200 points 76%, 300 85%, 400 92% etc. (You can check these figures here if you'd like).
I was reading a blog of a master rated in the 2300s who would give knight odds at his club to class A players. He said at about 500 points difference is when the knight odds started to break even for him, meaning, the lower rated opponent could score about 50%. This means at approximately master level a knight for no compensation will drop their rating about 500 points. It was a long time ago but I may remember reading something about sub 1000 players who drop a knight effectivly loose only 50 rating points.
Matej Guid and Ivan Bratko compared the top players of all time by analysing their best games with Rybka and finding the % of error according to Rybka. There are many many breakdowns of the statistics they gather, but the most accurate player (considering all the selected games for that player, not the single most accurate game for example) they concluded was Capablanca. You can view it here.
What's interesting is it shows even the "worst" world champs blundering less than the worth of a pawn over an entier game -- also their moves agree with Rybka sometimes as much as 70+% of the time! (I mean really, you might as well be playing a CPU). For such world class players I'm sure even pawn odds would drop thier rating hundreds of points.
Has anyone had a similar experiance as the master who gave knight odds? How many pawns do you think you'd have to give up to loose to someone rated hundreds of points lower than yourself? Let me know if anyone here has come across something similar. It would be interesting to have a relation between material and rating, or, how much a pawn is worth at each class of player.