As with most performance questions the best way to find out is to Benchmark it:

use Benchmark qw(cmpthese);

cmpthese(-5, { NoPlus => sub { $_ = 'a' . ('x' x 1000) . 'b'; s/x//g }
+,
               Plus   => sub { $_ = 'a' . ('x' x 1000) . 'b'; s/x+//g 
+} });
[download]

Produces these results:

$ perl subs.pl
           Rate NoPlus   Plus
NoPlus   2674/s     --   -98%
Plus   115762/s  4230%     --
[download]

So it would appear that using + for your iteration is 42 times faster than using s///g! Of course the exact difference will depend on the string being matched. When I changed the number of x's to 100 the difference dropped to 12 times. When I changed it to produce a string of 'axaxax...' there was almost no difference between the two solutions.

-sam