Horses for courses I guess - purely threaded vs purely MP both have different set of tradeoffs and shoehorning one over the other always fails some use cases. The article says they are also considering the possibility of having to keep both process and thread models indefinitely for this and other reasons.

I know nothing of PG internals but I can see why process per connection model doesn't work for large machines and/or high number of connections. One way to do it would be to keep connection handling per thread and still keep multiprocess approach where it makes sense for security and doesn't add linear overheads.