When you go to a chess tournament, why is your rating calculated by your tournament performance instead of how most internet chess servers do it? Why don't they adjust the rating by game? What is the advantage.
The way that rating systems work, your most recent results have a greater influence on your rating than your earlier results. So if your rating was adjusted after each game in a tournament, then the games in later rounds would affect your final rating more than the games in the earlier rounds. Most of the time this wouldn't make much sense, as a tournament usually takes place over a fairly short period of time. It makes more sense to adjust the rating just once, after the tournament has finished, with all games counting equally.
Several reasons. First of all, it's necessary to assign some sort of rating to any unrated players in the event, so that their opponent's rating can be calculated properly. So a quick "pre-calculation" of the performance rating of the unrateds gives us an approximation we can use to calculate their effect on their opponents' ratings. That can't be done well without a tournament's worth of games. So if the chess servers aren't making modifications to everyone's rating in every previous round as well as the current one, they're doing it wrong, and are getting bad results (GIGO). And, once you start doing that, you upset the pairings that were made based on those ratings.
Also, not every tournament can spare the time to calculate ratings and re-order the entire tournament table every round. (You do realize that the rating order of players affects every round's pairing, right?) Quite often there's already a scramble between rounds to get everything ready to go for the next without doubling the work. (Yes, I've organized and run tournaments, how could you tell?)
And no tournament software exists to do that outside of the chess servers. None of the packages TDs normally used can do that. (And the chess server software isn't approved for use in an OTB tournament, AFAIK.)
The only reasonable way around those are round-robin pairings, but they are not an option for large events.
There's lots of new and interesting ways to apply technology to events, but remember, once you require a certain level of technology, you lock out everyone who doesn't have it (or doesn't have the money or inclination to get it). There's still some players who refuse to play in events with a delay clock -- "30 minutes should MEAN 30 minutes, not some unknowable/incalculable time between 30 and 40 minutes!"
I've worked with folks who could set up the last round of a 100-player swiss using index cards in mere (as in 5 or less) minutes. Forcing folks like that to buy a computer and software just to run an event would never fly.
And that, in the final analysis is why it's not done that way. We can have time controls with and without delay, and let the people select the events they participate in. But every player participates in the rating system, so we can't have some events rated one way while others, played under identical conditions, get rated a different way. I guarantee hard feelings will result.
I think the major reason is historical - it was easier to calculate in the pre-computer era. All you need to do is, for each player, compute rating average of the opponents and look up one value from a table. Also, making a miscalculation, you only need to re-compute the rating of one player, while in the other model one mistake could affect the whole thing.