It seems, at least in the case of NTFS and ISO9660, that the filesystem is actually more of a bottleneck than the disk itself. Lots of small files can cut the transfer speed down by 2-3 orders of magnitude, even down to about 100 KB/sec. Linux filesystems seem to do much better on small files, but it makes me wonder what particular access strategy is causing the bottleneck.

The data should even be sequential on disk, making this all the more puzzling.