"If I find just one experienced OTB and chessclub player that tells me that the rating difference from 1400 to 1450 represents real playing ability difference in all cases , I will be very surprised."
Well, I'll be one of those people, then :)
I guess your point is that lower rated players are much less consistent in how they play. I could even argue against that one, since some players are much more consistent and others are much more wild and vary a lot more in the quality of their games (they might play like 1900 one day and another day 900). But I guess you're referencing how much more often there are draws at the top level, things seem stable, so I'll grant you this point for now.
But if you have a 1400, and a 1450, who both change ratings a lot, it definitely takes more skill to average out at 1450 than to do so at 1400. It wouldn't make sense for two players of equal strength over the course of years to continue to have an average rating difference of 50 points. The average should be the same over a large amount of games, if neither player improves.
Djonni, the average USCF rating is 1068.
http://archive.uschess.org/ratings/ratedist.php
FIDE doesn't make finding out the average rating easy. I don't know if it's published anywhere. Of course, it will almost certainly be higher than USCF if only because it's a bit more difficult to get a FIDE rating. That means that only the most competitive players will have FIDE ratings. Once upon a time, the lowest FIDE rating possible was 2000. Now, the rating list is much more open.
We are 97 rated players in my club. No 48 is at 1647 FIDE. In the last Tournament we arranged open to all fiderated players in the world we were 162 players and the 81 rated was ca 1500, so it looks like I estimated to high normalrating. A lot of the kids was not rated, and they are counted here among the 162.