It's not just performing HLR, it is creating records representing all the real and silhouette edges, both completely visible and partially occluded.
Identifying the edges could be done on separate cores, but each core is competing for access to the same in-memory database to keep the edge list current. It probably could be done, but writing parallel applications is easiest when there is little interaction, such as ray tracing where each pixel color is independently arrived at.
On the way to slowdom - I've had users import shrink wraps of the top assemblies into sub-assemblies to make installation drawings, with same sub-assemblies going into the assembly the shrink wrap came from. Query select an item and it turned out there were 2-4 copies of the top assembly surface geometry in the top level. If it's blanked it ceases to exist, right?