Hello,
I cannot speak for what Google is doing, but among GTFS producers and consumers it has been common to distribute GTFS-RT messages incrementally/differentially via message queue systems for over a decade. The technologies in use may not be the ones you mention, but the principle of sending small messages with low latency rather than repeatedly fetching a file is there.
This practice is not very well documented, but there is a Github issue about updating the spec where I have collected some examples of GTFS-RT producers using this approach and some characteristics of their systems:
Regards,
Andrew