Looking back up the thread, I see plenty of hyperefficient
solutions to this. Well, here's my shot at the least efficient
solution... apologies to princepawn who probably isn't aided
in the least by this post

sub fisher_yates()
{
  for(my $i = @_; --$i)
  {
    my $j = int rand ($i+1);
    next if $j == $i;
    @_[$i,$j] = @_[$j,$i];
  }
  return @_;
}


sub letter()
{
  for(@_)
  {
    my $k = int(rand 52) - 1;
    return chr $_[$k] if($_[$k] eq $_);
  }
  &letter;
}

# &letter isn't completely random...
# print(&letter(65..90,97..122),"\n");

# ...so use Perl Cookbook's Fisher-
# Yate Shuffle to randomize the list
print(&letter(&fisher_yates(65..90,97..122),"\n");
[download]

Update: Running above code several times has
confirmed my initial suspicion: &letter() as written is
more likely to return an 'A' than a 'z'. Fixed this by adding
a Fisher-Yates shuffle; this helps fulfill my original goal
of writing the least-efficient algorithm for returning a
random letter...