The problem you're likely having is that the workflow activity data is building up with each iteration of your loop which bogs down performance and makes the workflow run progressively slower.

I ran into this with my very first Laserfiche project in which I was routing tens of millions of documents and I found that the more activities in an instance, the more data, and the slower it got over time.

For example, when I'd kick off the workflow it would take less than 10 seconds per document but after thousands of iterations that would climb to minutes per document.

I'd be willing to bet that if you look at your instance details you'll find that the recent activities are taking much longer than those at the start and is this because of how much data has accumulated.

To maintain consistent performance, the best practice is to break this into chunks by changing your workflow do this in smaller batches and call that workflow from a parent process until you're done.

By breaking it into batches like that you ensure a fresh workflow instance for each batch and that way the activity data doesn't pile up so much that it negatively impacts performance.

It will depend on the type and number of activities, but as a general rule of thumb I try to keep my workflows to a maximum of 250-500 iterations per instance.

What you could do is look at your current instance to see at what point the activity durations started to get too bad and use that as your cutoff point.