You can use multiprocessing to parallelize the downloads and conversion to Arrow by passing num_proc= to load_dataset.
num_proc=
load_dataset