Assuming the servers support range requests. Could download `n` chunks of each file and then merge the results afterwards.