Graphile Worker maintainer here; keep in mind that postgres is not the ideal location for a job queue, so you’re going to be limited ultimately by postgres’ capabilities. I’ve seen Worker max out at around 10k jobs/second but very much YMMV - you should benchmark it for your expected use case. Personally I’d move to a dedicated job queue if I started having an average of more than 1-2k jobs per second. The majority of systems never hit anywhere near this (we very much cater to the “long tail” of job queue needs).
Regarding the horizontal scalability; that relates to if you have heavy tasks (tasks that take a second or more to execute) - you can use more instances to get higher throughput.
Hope this helps!