Right, giving away too many labels at once would make it too easy for people to copy the map data. By requiring much higher zoom levels, people would have to do many more requests to grab everything.
Not applicable at the zoom scale used in the article, but on higher zoom levels this is certainly a factor.
There might be something in that, yes. Obviously I don't know the details of Google's internal tech, but certainly that's true of the OSM equivalents: Mapnik (produces raster tiles) is much better at label placement than Mapbox GL (renders vector tiles).
Do you mean they're focusing more effort on reducing the complexity of the maps, to save bandwidth, because vectors derive much larger bandwidth savings from reduced complexity than bitmaps would?
Not applicable at the zoom scale used in the article, but on higher zoom levels this is certainly a factor.