Something similar has already been tried for the next version, so I guess choosing the fastest sources on-the-fly while downloading will be implemented.