Karl, good catch. Given the statistics of the corpus that I use (~ 350 directories compare to ~ 65,000 files), I didn't expect much difference, but I did get a surprise. Adding $rule->file made PIR (both iter and iter_fast) much slower than the alternatives! (FYI, this is using the Benchmark::Forking version of the program).

                  Rate  PIR PIR fast File::Find perl readdir find pipe iter find pipe
PIR            0.622/s   --     -36%       -67%         -77%           -88%      -89%
PIR fast       0.975/s  57%       --       -48%         -64%           -82%      -83%
File::Find      1.89/s 205%      94%         --         -31%           -64%      -67%
perl readdir    2.74/s 342%     181%        45%           --           -48%      -53%
find pipe iter  5.32/s 757%     446%       181%          94%             --       -8%
find pipe       5.80/s 833%     495%       206%         111%             9%        --

Cheers, lbe