No, it's not the extra weight of the subroutine calls, although sub calls do cause more overhead. What's causing this is what I mentioned earlier: you have scoping problems. None of the code you're benchmarking is actually using a valid value for either %url or $now. So your results are completely bogus.

The valid results are those that use a properly scoped variable. Aighearach's code did this, though I'm not sure it did so intentionally.

You can accomplish this by declaring your vars as package globals:

    use vars qw/$now %url/;

    $now = 8;
    %url = ( monday => { @{[map(($_,1), (1..1000))]} } );
[download]

This gives valid results.

Benchmark: timing 1000 iterations of Grep, Max, Ternary...
      Grep:  6 wallclock secs ( 5.49 usr +  0.00 sys =  5.49 CPU)
       Max:  5 wallclock secs ( 5.81 usr +  0.00 sys =  5.81 CPU)
   Ternary:  7 wallclock secs ( 6.42 usr +  0.00 sys =  6.42 CPU)
[download]