pyarrow not effective at speeding up toPandas() on large dataset

Question

I am using python 3.7, pyspark 2.4.5, pyarrow 1.0.1, pandas 0.25.1 in Windows, with the environment variable ARROW_PRE_