That speeds up parameter passing by removing the need to make a copy.

That's only part of the story. It's also, even most of all, the way HTML::TokeParser distinguishes between literal HTML in a string, and a string holding the file name of a file containing the HTML:

$p = HTML::TokeParser->new( $filename );
$p = HTML::TokeParser->new( $filehandle );
$p = HTML::TokeParser->new( \$document );
[download]

Note cases 1 and 2.

Choosing this convention instead of the other way around, was indeed the wisest choice with regard to efficiency.