P.S. to the above post: grep should be slower than the other two algorithms. Aighearach's results were predictable, where mine are distinctly opposite...
Update:
Having seen a few posts, I have to "update my version of reality" (as a friend always chides me for doing slowly). The grep sort implementation is running faster on everyone else's machines, too. Algorithm analysis would conclude that since sort is in O(N log N), it should be slower than an O(N) algorithm (Max and Ternary should both be in O(N)). Real-world results don't show this.

What's going on? Somebody help me update my version of reality! :-)

Here is what happens on my system:

Benchmark: timing 1000000 iterations of Grep, Max, Ternary...
      Grep: 10 wallclock secs (10.75 usr +  0.05 sys = 10.80 CPU)
       Max: 31 wallclock secs (31.10 usr +  0.05 sys = 31.15 CPU)
   Ternary: 32 wallclock secs (31.29 usr +  0.01 sys = 31.30 CPU)
[download]

It is a dual PPro 200 (180, overclocked), RedHat 5.2 (seriously upgraded - 2.2.12 kernel).
perl -v says:

This is perl, version 5.005_57 built for i686-linux-thread
[download]