If it's not partitioning data per processing unit then it would not be considered a "thread-per-core" architecture based on the definition the article provided. Work stealing means more than one thread can be responsible for a single piece of data. This could result in crossing NUMA nodes or servers.