Unexpected datasize issue with snappy compressed parquet files

Question

df1 - large dataset
df2 = df1.sample(tiny_fraction)

df1 is written to disk as a parquet with snappy compression (~75GB)
df2 is written to disk as a parquet with sna