Assuming that this is a utility script to globally update a web site, and will not be run by visitors. I would go with your first inclination and use File::Find to walk the directory, and complete the operation in one call to the cgi script. Just print some feedback to the browser (even a single space should do) after each file update and you will not have to worry about timing out. I've used this technique often for long running cgi utility scripts. Just be sure to print a completion message so you know when the job is done.

use POSIX '_exit';
use File::Find;

my $search_folder = '/some/folder';

find sub {
    return unless -f;
    local @ARGV = $_;
    defined( my $pid = fork ) or die "Cannot fork: $!";
    unless ( $pid ) {
        local $^I = '';
        while ( <> ) {
            s/some text/another text/g;
            print;
            }
        _exit( 0 );
        }
    }, $search_folder;
[download]

Do you want to modify FILELIST.txt, or the files that are listed there?

If you want to modify only one file, use Tie::File.

But maybe you could tell us the underlying problem, perhaps there's a better solution to it.

I want to modify FILELIST andfiles that are listed there.

In order to continue from previous program execution i need to know there the program stop.

I want to modify FILELIST andfiles that are listed there.

In order to continue from previous program execution i need to know there the program stop.

Yes. The buyer say that CGI exucution time is limited. Also i want to insure that i can continue searching if something bad happens (abnormal program termination).

I'm not getting a picture..
A limit of the execution time makes no sense to me..?

If you've to deal with the browser timeout (I didn't formulate my question exact enough) I could think of a script which runs as a server, independent of the web server.
You could start|control|stop the replace script through the web interface.

I'd also say that you've more control over the separated replace script this way,
the replace script could be restarted by a cron script if it's killed,
oy you are able to run the script as another user as the web server/ with an other priority.

[find and replace script]<--IPC-->[webserver cgi sript]

The web interface could even update the progress and results through some javascript (ajax)

This is at least how I would do this.

beside this, storing the progress of the traversal in a filelist file seems like a good idea to me.