Build ferret for large-file support


I want to using ferret to do indexing and searching in my application. I
am indexing a large number of documents say, 5 million. What should be
done to optimize ferret for large file-support? Currrently the
:max_merge_docs value is set to INT_MAX. Would changing the
:max_merge_docs value help? I have changed but did not see a significant
improvement in performance. What else can I do?