Rank 1 (Script Task): Using a!forEach() allows for bulk processing in memory. This is the most performant method as it avoids the massive overhead of spawning multiple process instances, resulting in the fastest execution time and lowest memory footprint. Rank 2 (Subprocess MNI): The native Subprocess node is optimized for internal process orchestration. configuring it for Parallel MNI ("start at same time") significantly reduces execution time compared to serial methods, though it still consumes memory for $N$ instances. Rank 3 (Start Process MNI): While also parallel, the Start Process Smart Service typically incurs higher initiation overhead per instance compared to the native Subprocess node. It is generally less efficient for tight data processing loops than the Subprocess node. Rank 4 (Flow Loop): "Looping over" a node implies a manual flow loop (Gateway Loop). This is the worst option because it executes sequentially (slowest time) and causes the parent process history to bloat indefinitely (highest memory risk), leading to severe performance degradation.

Appian Documentation - Process Model Performance:

Section: Processing Large Data Sets

"Use the a!forEach() function... to process data in bulk. This is significantly faster than processing items one by one in a process model."

"Avoid using Multi-Node Instances (MNI) for large data sets if a bulk operation is possible."

Appian Documentation - Sub-Process Activity:

Comparison with Smart Services: The documentation notes that while Start Process handles load balancing, the Subprocess node is the native method for parent-child orchestration and generally offers tighter integration/performance for data flows.

Appian Playbook - Looping:

Identifies Manual Looping (Flow Loops) as a pattern to avoid for large lists due to "Process History Bloat" and recommends replacing them with MNI or Bulk Processing (a!forEach).