Firstly, chess.com uses the Glicko rating system, not the Elo rating system (see: http://en.wikipedia.org/wiki/Glicko_rating_system).
The idea of the initial games being worth more or less, has solely to do with the RD (ratings deviation) value. The idea is that as the player progresses, their RD decreases, which means there "less deviation" in their rating (thus making it "more accurate"). Even if an 1800 player plays through, and wins a ton of games, with a new account, he's still going to lose his game against the 2000 rated player (most likely). The idea is that as the player plays more and more games, his rating is more and more accurate. Eventually, the RD will be so low, that he'll even out and his rating should ACCURATELY represent his skill level.
Yes, there definitely is the inflated looking ratings sometimes, but that's just due to the person having few games under their belt. You can see I have over 1300 games on my live blitz chess, so my live blitz rating (1371) is probably fairly accurate (notice it's only a few points above the starting score of 1200, and my highest was 1478. Interestingly, my 1478 was actually acheived around the 900th game, I believe.)
Just remember that if a player loses when his RD is high, he also loses a LOT of points (more than he would've gained for a win), and that's what makes things work.
Someone I know who I got to start playing on Chess.com a couple years ago and who quickly reached a rating of 1900+ got to the point where games were worth a very very small number of points. I know how many points you gain/lose for winning/drawing/losing depends on your own rating as well as your opponents rating relative to yours but at the time I thought the reason this guys games were worth so little was because he was rated so high and that's just part of the system.
I now realize that the # of games played is also a very big factor because my games while I'm rated in the 1100's and I'm playing against people between 1100 and 1300 tend to allow for 5 to 30 or so points to be won in a game but someone who just registered jumps up and down by like a hundred at a time when him and his opponents are in the same rating range.
This isn't how the ELO system works with FIDE, is it? Is a players ability to win/lose points directly and significantly affected by his number of rated games played regardless of what his current rating is? If not, wouldn't it make more sense if Chess.com's rating system wasn't like this? In it's current form, aren't players who have played a lot on this website kind of penalized for how much they've played? Let's say someone has played many thousands of live games and there comes a point where he starts taking chess seriously or gets a really good instructor or something and he improves substantially over a period of time and is winning against most similarly rated players and wants to get his rating up. Does the current system not then give him an incentive to make a brand new account and start fresh at 1200 then work his way back up to his old accounts rank and beyond much faster and easier because of how many more points each win gets him with his new account?
If I'm mistaken about the way it's currently calculated, please let me know. If there's a good reason that it's like this, I'm curious as to what that reason may be. If there's no real good reason for this, why not give people who have a ton of rated games under their belt the same opportunity as a player who just registered yesterday?
I'm asking just out of curiosity and humbly suggesting this while acknowledging the fact that I don't really know much about it so my perspective may be missing something crucial. I'm not really posting out of concern for my own rating. I haven't improved substantially recently, don't put that much time/effort into chess and would rather struggle to raise my own rating through the 1100's the way I am now than to have it rapidly & severely fluctuating in the 1000-1300 range as it would if each game was worth 5-10 times more points won/lost. I just don't understand why as your Chess.com record grows your games should be progressively have less and less of an impact on your rating. I see how naturally they would have less of an impact on your average overall stats because each game would be just one of an ever growing total recorded, there's nothing wrong with that, but the the way the skill rating is calculated seems heavily affected by something that does not logically correlate to skill, and that's what I'm confused about.