When to avoid? When using deduplication storage, it is recommended to disable the "Defragment and compact". As deduplication appliances are fragmented by their very nature, and have very poor support for random I/O workloads, the compacting feature will not enhance backup or restore performance.
What would be the recommended option? I'd be glad to hear some recommendations from the field about enabling this option with Data Domain
Shoud per-vm backup file chains be considered when deciding whether to enable compacting or not?
I believe the recommendation to enable this option for dedupe appliances was given prior to introduction of per-VM backup chains, when reading metadata could take significantly longer in case of large highly fragmented files. However, compact operation is pretty heavy random I/O and can itself take considerable time, hence recommendation from the best practices guide. Anyway, it's better test in your particular environment and derive recommendations from there.