>>I see you've discovered that each WP is multi-threaded. The simplest way of avoiding the dictionary rebuild problem might be to put a lock around it.....
>
>There is already a lock. The initial locking mechanism, as discussed a few months ago, also applies to the rebuild situation when needed. The issue is about an already ongoing hit in progress from the same worker process. Thus, that hit might loss some of the application functionality. A workaround I have found is to put a timeout of a few seconds before the reload takes place. So, this should leave enough time for such an already ongoing hit in progress from the same worker process to complete.
>
>Any comment on that approach?

Maybe I'm missing something you've already explained, but what if new hit(s) come in during your timeout period?