As you try to do that, if you happen to notice that the self-contained test script (using the same test data in the same way) actually runs in about the same time on both systems, then that would suggest there might be a problem with where your linux server is getting its Japanese data from (i.e. it might not be a problem with your perl script or the perl interpreter).

(updated to fix grammar)

It's really, really long code that I'm inheriting, so it's difficult to post a small self-contained portion, but those regexs I posted should cause the problem on their own when run against certain lines of Japanese. An example of one of these is: "インドネシア休場"

I cannot reproduce that. In fact, according to Benchmark, 5.8.8 is significantly faster than either 5.10.1 or 5.12.2.

#!/usr/bin/perl

use strict;
use warnings;

use Benchmark qw[cmpthese];

my $RUNS = shift || 5;

print $], "\n";

sub match {
    my $subj = shift;
    $subj =~ /^\s*SPECIFIC\sDECISIONS\s*$/; 
    $subj =~ /^\s*Important\s+\S+\s+Decisions\s*$/; 
    $subj =~ /^General\s+Decisions$/; 
    $subj =~ /^E-mail:\s*replies\@x.com$/;   
}

for (1 .. $RUNS) {
    cmpthese(-3, {
        english  => sub {match "This is a string"},
        japanese => sub {match "インドネ&#1247
+1;ア休場"},
    });
}

__END__
5.008008
             Rate japanese  english
japanese 466031/s       --     -38%
english  751158/s      61%       --
             Rate japanese  english
japanese 505266/s       --     -36%
english  793800/s      57%       --
             Rate japanese  english
japanese 502207/s       --     -36%
english  778988/s      55%       --
             Rate japanese  english
japanese 498022/s       --     -35%
english  767420/s      54%       --
             Rate japanese  english
japanese 508507/s       --     -35%
english  780966/s      54%       --
5.010001
             Rate japanese  english
japanese 339534/s       --     -41%
english  577872/s      70%       --
             Rate japanese  english
japanese 416203/s       --     -28%
english  577831/s      39%       --
             Rate japanese  english
japanese 369283/s       --     -37%
english  581775/s      58%       --
             Rate japanese  english
japanese 418431/s       --     -19%
english  516905/s      24%       --
             Rate japanese  english
japanese 417789/s       --     -29%
english  587599/s      41%       --
5.012002
             Rate japanese  english
japanese 361165/s       --     -27%
english  496107/s      37%       --
             Rate japanese  english
japanese 372533/s       --     -33%
english  553731/s      49%       --
             Rate japanese  english
japanese 362232/s       --     -34%
english  549086/s      52%       --
             Rate japanese  english
japanese 361565/s       --     -33%
english  537911/s      49%       --
             Rate japanese  english
japanese 368166/s       --     -32%
english  539974/s      47%       --
[download]

(For some reason, if you have "インドネシア休場" in a a <code> block, Perlmonks escapes the Japanese characters, and then shows the raw entities. Bug?)

You said:

It's really, really long code that I'm inheriting, so it's difficult to post a small self-contained portion, but those regexs I posted should cause the problem on their own when run against certain lines of Japanese.

One step in the diagnosis is to determine whether it really is the regexes themselves (and their handling of Japanese), or whether it is instead some other problem involving the Japanese data in the big app, regardless of the regexes (e.g. only the Japanese data are coming from a particular source that isn't playing nice with the app).

You have a local machine and a remote machine; you have a small sample of Japanese text and non-Japanese text; you have a set of regexes. If a single minimal test script, containing its own test data and applying those regexes, runs in roughly the same amount of time in both the local and remote machines (or at least, behaves consistently when applying the regexes to the Japanese and non-Japanese strings), then the problem in the big app is probably not being caused by the regexes.

Here is what it boils down to: What sort of evidence did you have that led to your conclusion in the OP that the regexes were causing the problem? And what is the likelihood that your original evidence might also be due to some other cause, not involving the regexes?

If you try the test script that JavaFan supplied above, and it turns out to show a really big slow-down on the remote machine, then you really do have a problem using those regexes with Japanese text on that machine. That would be strange, but in that case, you can try some different variations on the regexes to see if there's another way to do the job without taking the hit.

OTOH, in the more likely case that the regexes are performing as expected in that test, you'll need to examine what else is going on with the Japanese data in the big app on the remote machine (and how it differs from the local one)...

Sorry, I mean that I'm using those regexs to match against a lot of different documents. Many of those documents happen to contain different languages. None of the characters in any of those languages cause any problems except the Japanese ones, which slow down the matching to a crawl (and it seems to only be about 10% of Japanese strings I've tried to match against that cause the problem). Does that clarify things?