I'm not a Perl guru, but I have been knocking around computing for several decades. My experience has been that I/O buffering, especially read i/o buffering, is managed by the o/s, via the drivers. Again, from my experience, there is fairly little you can do at the application level to manage this.

You could, in theory set $/ to a (fairly large) numerical value, and see if it makes a noticeable difference, but I suspect the overhead introduced in explicitly processing end-of-record markers would eat up any savings.