Ideas and suggestions.

Conclusion: the K coefficient is unrealistically high (not only G+U compared to others, but also in general for all templates)

Quite a strange conclusion. The coefficient is simply multiplied by all rating changes, so if the K coefficient is reduced by half for all templates, then the rating growth rate for all players will also decrease by the same factor of 2. In essence, one rating point will be worth two now. It will be equally difficult for newcomers to catch up with the top players, whether in the case of the top players gaining twice as many rating points with the current system, or in the case of the current rating values of the top players with a system with a K coefficient reduced by half.