These integrations must be optimized for files and disk I/O rather than messages to mitigate the decrease in performance caused by over using memory. When payloads are maintained within the message size constraints, then it is safe to use message objects to manage the data during the integration runtime and avoid file I/O. The guidelines are (mileage may vary depending on the server environment and configuration):

The datasets tend to grow over time. Also, the sample sizes you work with when you originally built the solution may be much smaller than the production environment in which they will ultimate run, so you must account for this possibility.

If you can predictably use djMessage objects for a transformation map source or target connection, use a sample file during the initial design session and then replace it with the djMessage URI scheme prior to packaging the map for deployment (for example, djMessage:///myMsg).