If OUTPUT, CLUSTER(<target-cluster>) is used and the target-cluster is wider then the cluster the query is running on, the compressedSize property will not be published.
This is because Thor always writes a part per slave (no more), so when a file is targeted and published to a larger cluster, it has to pad out the file width with empty parts.
However, it does not fill in the meta data for those parts.
And when calculating the sum total, it stops if some of the part @compressedSize info. is missing.