By now I have found kind of a workaround myself. To get more timesteps for the animation, I simply slowed down the background footage (say to 25%), did the tracking and put in the fire. Then I discarded the background footage, sped up the fire animation to the according 400% and put the original footage in the back for composition.
Not the most elegant or fast way, but it worked for me to get 4 times the amount of simulated frames compared to the original, where only every fourth is written out in the end. It was enough for me to get a smooth track when applying motion blur in addition.
If I find a better way, I will report back.
Greetings from Germany,
Floh