When using multistreams, requests are sent in parallel. curl keeps
track of different time values for each handle independently. This
means that we can retrieve the transfer duration for each handle
separatly.
It is then possible to calculate the average time taken by each
transfer independently. Even so this might be an interesting data to
compute, this does not reflect the benefits of using parallelism.
Therefore the average time taken by a request is calculate as the
total time elapsed to perform all the requests divided by the number
of transfers.