@vladimir_u in the tests we did using a E3D V6 hot end, the compensation needed to act instantly, because the additional heater power took a small number of seconds to reach the nozzle, whereas the fan took less than a second to reach the new speed. So the fan speed increased faster than the heat reached the nozzle. As a result, the best compensation we could achieve still resulted in a slight dip in indicated temperature when the fan speed increased; but this dip was smaller than without heater feedforward, also the recovery to the set temperature was faster.
I am surprised that in your graph, the temperature is rising so much because of heater feedforward. Have you tested whether the heater feedforward is compensating correctly when you change the fan speed and leave it at the new value for several seconds? It may be that the feedforward constant (the second K parameter in the M307 command) is not appropriate for your hot end.
What type of hot end do you have, and what type of print cooling fan?