Presently there is no option for a customer to specify where to store attachments when using Confluence Clustered version. It is by design decided to store the attachments in Database only, due to synchronizations issues and tangosal distributed cache, I trust.

I would like to propose two approaches to allow customer to opt for the way attachments should be stored.

1. Store only the footprint of the attachment path to the actual File System in the DB. Hence, when application querying a particular attachment, it will retrieve the footprint (absolute file path) and fetch it from it's location.
2. Allow to store attachment in a shared filesystem and ensure that all the copying between nodes is synchronised.

I have no doubt that the above suggested changes are not 5 sec fixes and would require appropriate effort being applied