From what I've read the swarm works better on one large file vs many smaller ones. I am verbatim passing along what I've read and have no data to substantiate that claim. Hopefully someone that has a better understanding of the protocol and client implementations can weigh in.