If destination disk is an SSD, just keep in mind that SSD's are, for what's in concern here, damn fast. (*)
If destination is anything slower (from a flash with not-so-high-write-rate to a SATA3 HDD to a magnetic tape) you should queue the transfers, one at a time.
Your bottleneck will be on destination disk for sure
In HDDs this advised nowadays because, even though there is more than one head in a cylinder, they are not allowed to read nor write concurrently (it has shown no performance impact in real world scenario due to the fact that we can encounter random data in one of the heads that would not allow us to R/W in that head for most of the time being, and if the heads are not completely centered that could lead to disk corruption).
Common sense (and mathematics too!) say that if we have a very very source but we can make the transfer of files in parallel the head seek could (in not many scenarios though) be faster than serialization of the job.
(*) That is, as nnewton states, because there is no head movement. But should we not use the term latency: latency is head movement + disk rotation delay + read time, and in SSD that is simply random read time and does not depend on the previous read position as disk drives do
How much is a "Bunch" – Moab – 2011-03-04T00:38:22.710
20+, question updated. – fredley – 2011-03-04T00:39:32.457
6There are some very interesting theoretical answers here, but I'd love to see some real data. How could we set up a reliable test? – Pops – 2011-03-04T05:00:22.303