I'm not sure what function the 100 has in the following. It doesn't seem to affect the resulting ratings.
They agree the game is worth 100 points
If A wins - A gets 100 (+25 now 2025) B gets 0 (-25 now 1475).
If a draw - A gets 50 (-25 now 1975) B gets 50 (+25 now 1525).
If A loses - A gets 0 (-75 now 1925) B gets 100 (+75 now 1575).


But that aside, it's an interesting idea.

I'd start, though, by saying that it's not a rating system. The purpose of ratings is to be able to compare players (as accurately as the model will allow). The model should be self-consistent, ie. there should be no player input required or allowed.

By having players agree to varying amount that they "put up on offer", (or bet? lol) - is to have a currency system where the "rating" is more like a purse or wallet. That's actually an attractive idea which could work in parallel with real ratings. At TrueMoneygames you can play Backgammon for money. But they also have the concept of play money for those who like safe betting. I was watching a player the other day who had amassed an amazing $1,000,000 of this play money. (Given that games are $100 that's quite an achievement - very aggressive doubling seems to have been the key. He won 3200 in a single game because of this. But, lol, I digress..) The point is that risking a chosen amount of your points is fun but cannot be part of a rating system.

I agree with you that a smaller K factor for tournaments could help encourage higher rated players to join. I haven't given much it study in Chess because I'm a Backgammon man. I do believe that the best way to encourage Backgammon players to risk their rating in an open-to-all tournament is to use the proper formula, one which is fair to all players.