This worst-case performance can happen if the list is already in is in sorted order and the pivot is picked by choosing the first element of the list as the pivot.
In my data structures and algorithm analysis class,
they told us to select one of the elements at random
to use as the pivot. This adds a small amount of
overhead (the amount of time needed to pick a random
number each iteration) to the average-case scenerio,
but it basically eliminates the worst-case scenerio,
effectively transforming it into an average-case
scenerio. "random number" here can be anything that
can pass as random. If your system clock has good
enough precision, you can just grab that. The
key thing is that you won't be picking the same
element every iteration -- sometimes an early
element, sometimes a late one, sometimes a middle
one. So it makes no real difference how the list
is sorted initially.
This is of course all moot now; these days we just
use Perl's built-in sort.
$;=sub{$/};@;=map{my($a,$b)=($_,$;);$;=sub{$a.$b->()}}
split//,".rekcah lreP rehtona tsuJ";$\=$ ;->();print$/