I second the suggestion of using HTTP refresh. It was the only way I found to reliably update data on various browsers (IE being particularly crappy about server-push).

My strategy is to spawn a background process that does the time-consuming work while redirecting the browser to a progress page. The background process periodically writes to the progress page. You can even change the refresh time in the meta tag on the progress page if you have an idea of how long you expect different steps in the processing to take.

A tip I can pass along about doing this, when forking off a child process in a cgi script, make sure to close STDOUT and STDERR in the child or the browser will wait for the child to complete.