For me, for is faster on one Perl, and slower on another. The lesson learned is that optimizations like this is version dependant and not something you should bother about when trying to speed up your program.

Benchmark: running tp_for, tp_map for at least 2 CPU seconds...
    tp_for:  2 wallclock secs ( 1.81 usr +  0.24 sys =  2.05 CPU) @ 11
+155.46/s (n=22891)
    tp_map:  3 wallclock secs ( 1.96 usr +  0.23 sys =  2.19 CPU) @ 10
+615.60/s (n=23280)
          Rate tp_map tp_for
tp_map 10616/s     --    -5%
tp_for 11155/s     5%     --

This is perl, v5.8.0 built for MSWin32-x86-multi-thread
[download]

Benchmark: running tp_for, tp_map for at least 2 CPU seconds...
    tp_for:  2 wallclock secs ( 2.02 usr +  0.28 sys =  2.30 CPU) @ 35
+695.65/s (n=82100)
    tp_map:  2 wallclock secs ( 1.80 usr +  0.30 sys =  2.10 CPU) @ 39
+710.48/s (n=83392)
          Rate tp_for tp_map
tp_for 35696/s     --   -10%
tp_map 39710/s    11%     --

This is perl, v5.8.4 built for i486-linux
[download]

ihb