Spark, why I'm getting better performance with count in the code

Question

I\'m quite new to Spark so I\'m asking this question here as I can\'t find a reason why my code behaves like this... In general I have a Jupyter notebook with number of cells wh