Changing maxIterAtScale effect performace?

[pborwarnginn]

I've noticed for each scale the maximum of iteration is different. I have reduced and made all scales to be equal. The results turned out to be nosier than the default config.
Would there be any efficient ways to adjust these numbers and train faster without losing the performance?
At the last scale is 200000, I assume it's the training data size n= 200000. If my data is less than 200000 should I change my dataset size?