Glad you enjoyed the article, part 2 will be available tomorrow which shows more technical information.
However to answer your question, the inclusion of an HDI Region (whether on-premise in the appliance or azure) should be seen as the addition of hardware and software and is a separate engine from the PDW. The data movement between HDI and the PDW regions is a fully parallelized process controlled by Data Movement Services (DMS) but PolyBase is still responsible for the predicate push down capabilities which uses MapReduce to reduce the amount of data moved between the two regions based on the query predicates.The final result set produced by the MapReduce job is then moved using DMS to temporary tables in the PDW.
Important to note though, this only happens when the amount of data that would be landed in the PDW region exceeds 1GB per distribution (and you have enabled the functionality by providing the resource manager location in the external data source configuration)
I just want to credit a friend of mine and an APS guru James Rowland-Jones for provide the specific details and exact context to this answer.
I'll also refer you to a very good article by another APS Guru James Serra : PolyBase Explained[/url] where he explains exactly how the current implementation of PolyBase works and its limitations.