NB: I doubt this feature has been used.
When writing to multiple non-local cluster, with e.g. OUTPUT,CLUSTER('cluster1', 'cluster2'),
and if those clusters are larger than the local cluster (the cluster the query is running on),
then every other cluster in the target list will not have empty file parts created for parts>local size.
As a consequence, reading back these files from those effected target clusters, would result in part not found type errors.