Let's benchmark:

#!/usr/bin/perl

use strict;
use warnings;

use Benchmark qw 'cmpthese';

our @dates = qw [092205 092305 092605];
our (@mb, @me, @fb, @fe);

cmpthese(-1, {  
    map_block => '@mb = @dates; map {s{(\d{2})(\d{2})(\d{2})}{$1/$2/$3
+}} @mb',
    map_expr  => '@me = @dates; map  s{(\d{2})(\d{2})(\d{2})}{$1/$2/$3
+}, @me',
    for_block => '@fb = @dates; for (@fb) {s{(\d{2})(\d{2})(\d{2})}{$1
+/$2/$3}}',
    for_expr  => '@fe = @dates; s{(\d{2})(\d{2})(\d{2})}{$1/$2/$3} for
+ @fe',
});

die unless "@mb" eq "@me" && "@mb" eq "@fb" && "@mb" eq "@fe";

__END__
             Rate for_block  for_expr map_block  map_expr
for_block 43115/s        --       -1%       -2%       -6%
for_expr  43530/s        1%        --       -1%       -5%
map_block 43840/s        2%        1%        --       -5%
map_expr  45948/s        7%        6%        5%        --
[download]

Running it a few more times shows the expression form of map being slightly faster than any of the others, and even the block version of map not doing too bad:

             Rate map_block  for_expr for_block  map_expr
map_block 43050/s        --       -1%       -5%       -7%
for_expr  43631/s        1%        --       -4%       -6%
for_block 45271/s        5%        4%        --       -2%
map_expr  46419/s        8%        6%        3%        --

             Rate  for_expr for_block map_block  map_expr
for_expr  42309/s        --       -3%       -5%       -7%
for_block 43530/s        3%        --       -3%       -5%
map_block 44660/s        6%        3%        --       -2%
map_expr  45583/s        8%        5%        2%        --

             Rate for_block  for_expr map_block  map_expr
for_block 43115/s        --       -1%       -3%       -7%
for_expr  43530/s        1%        --       -2%       -6%
map_block 44246/s        3%        2%        --       -5%
map_expr  46419/s        8%        7%        5%        --
[download]

Things might be different on your machine, but the difference in performance is too small to regard one as better than the other.

Now, if you really want to go for speed, your concern should not be the difference between 'for' and 'map', but the difference between 's///' and something else. For instance, let's introduce an additional clause in the benchmark:

pack => '@pk = @dates; @pk = map {join "/", unpack "A2 A2 A2", $_} @pk
+'
[download]

This leads to the results:

             Rate  for_expr for_block map_block  map_expr      pack
for_expr  42309/s        --       -0%       -2%       -9%      -48%
for_block 42309/s       -0%        --       -2%       -9%      -48%
map_block 43115/s        2%        2%        --       -7%      -47%
map_expr  46394/s       10%       10%        8%        --      -43%
pack      80736/s       91%       91%       87%       74%        --
[download]

It dwarves the difference between map and for.