The problem is not with vec, it is the way I start at the start of the string and count off non set bits till I reach $rand. The following timings are on a dev machine being used by a few so there is plenty of noise. The time to read the last bits of a long string using vec looks constant even with strings spanning several orders of magnitude.

#!/usr/local/bin/perl -w
use strict;
use Benchmark qw(timethese);
timethese(5000000, {
    '1K'   => vectest(1000),
    '10K'  => vectest(10000),
    '100K' => vectest(100000),
    '1M'   => vectest(1000000),
    '10M'  => vectest(10000000),
});

sub vectest {
    my $n=shift;
    my $s="";
    vec $s,$n,1;
    my $look=vec $s,$n-10,1
}
__END__
Benchmark: timing 5000000 iterations of 100K, 10K, 10M, 1K, 1M...
      100K:  0 wallclock secs ( 0.59 usr +  0.00 sys =  0.59 CPU)
       10K:  1 wallclock secs ( 0.42 usr +  0.00 sys =  0.42 CPU)
       10M:  1 wallclock secs ( 0.49 usr +  0.00 sys =  0.49 CPU)
        1K:  1 wallclock secs ( 0.89 usr +  0.00 sys =  0.89 CPU)
        1M:  1 wallclock secs ( 0.51 usr +  0.00 sys =  0.51 CPU)
[download]

I have a serious optimisation up my sleave but and rather too busy to code it now :( Hope to have a shot when I get home this evening

Cheers,
R.