How do you define realistic for chess ratings? Actually, they are all realistic. Ratings should be considered inside their own population because your rating in server X is your performance relative to other players in the same server. As it depends the population you shouldn't compare one rating to another rating in another server. But if you really want to compare your rating to another population approximately, please see this method: How does Chesscube.com rating compare to USCF and FIDE ratings?
All chess rating systems follow almost same rules, but they have different converging speeds. This is what you explained in your question, one server gives less on win and takes less on loss, another one gives/takes more etc.
There is a thing, K-factor, in USCF rating system for example, to tweak this converging speed from game to game. Please see this topic: What does "half-K" mean in the USCF tournament rating system?
When you enter a rating population as a new player, you start with average rating of that population(generally). But it doesn't matter which rating you start with, even if you start from 3000, you will converge to your real rating as you play games. So, it's not about being realistic again.
And generally for all rating systems, your first 8 to 20 games will have super high converging speeds, to approximate your rating faster. After that, converging speeds drops to normal and ratings get stabilized.
So, shortly, your ratings will be different in different populations, and it's totally normal.