Neither stand up too well unless I've screwed something (which as we all know is entirely possible :):

Update: Code corrected in light of ikegami's post below:

C:\test>byChar.pl
...
    rgx_scalar => q[
        our $string;
        while ( $string =~ /(.)/sg) {
            my $c = $1;
        }
    ],
    rgx_list => q[
        our $string;
        for my $c ( $string =~ /(.)/sg) {
            ;
        }
    ],


C:\test>byChar.pl
                Rate split rgx_list substr_refs rgx_scalar unpack subs
+tr chop rev_chop pre_split
split         9984/s    --     -34%        -74%       -75%   -76%   -7
+7% -82%     -83%      -94%
rgx_list     15104/s   51%       --        -61%       -62%   -63%   -6
+5% -73%     -74%      -91%
substr_refs  38242/s  283%     153%          --        -3%    -7%   -1
+2% -33%     -34%      -78%
rgx_scalar   39574/s  296%     162%          3%         --    -3%    -
+9% -30%     -32%      -77%
unpack       40959/s  310%     171%          7%         3%     --    -
+6% -28%     -29%      -76%
substr       43352/s  334%     187%         13%        10%     6%     
+-- -24%     -25%      -75%
chop         56695/s  468%     275%         48%        43%    38%    3
+1%   --      -2%      -67%
rev_chop     57962/s  481%     284%         52%        46%    42%    3
+4%   2%       --      -67%
pre_split   173576/s 1639%    1049%        354%       339%   324%   30
+0% 206%     199%        --
[download]