I have this scary feeling that Linux's swapfile management is still based around the idea of swapping out 4096-byte pages, in random order, and then swapping them in one by one as you realize you need them.  With no readahead or clustering whatsoever.