First, I didn't say it would be faster, just simpler. I don't know what you mean about being Byzantine and obscure. It's pretty straightforward to me:

use File::Find;

find(\&matches_regex, '.');

sub matches_regex {
   return undef unless (m/.+\.txt$/ && -f);
   unlink $File::Find::name,"\n"; ## scalar contains full path
}
[download]

This will find and unlink all TXT files in a tree. Now, that may not be as concise, but I personally found it easier to control and understand than readdir loops. YMMV, I suppose. If performance was the primary goal, I would unequivocally agree that File::Find is probably not what one wants.