A typical Unix system will read from the hard drive in increments of 4x that or 4096 bytes. Here the buffer size is twice that or 8192 bytes. It is counter-intuitive, but increasing the buffer size can actually slow things down if you have a smart disk system.

With respect to disk reads, it doesn't really matter what size you specify with Perl's read(). Perl uses its own internal buffer anyway, which is 4k (hardcoded in the Perl source, i.e. not configurable, except by recompiling perl). In case you don't believe me, do an strace on your sample code, and you'll see that the underlying system calls always return 4096, independently of whether you specify 1k, 2k, 8k, or whatever size with read().

But you could use sysread(), which is actually implemented in terms of the read(2) system call, and thus does pass through the size you request. Whether the latter actually maps to disk block read requests is still another story, though... (depends on the OS).

You might also want to read 4k read buffer is too small.