Just to sum up the comments above:
Adjusting the original example to Flux.from(req.receive().aggregate().asString().delayElement(Duration.ofNanos(1)).filter(str -> (str.length() > 0))).next();
resulted in being able to exceed the 500 connections per second.
Upon studying delayElement
the documentation states that the parallel
scheduler is then used.
Thus the constraint appears to be the scheduler. Once that theory was established it was tested by replacing delayElement
with .subscribeOn(Schedulers.boundedElastic())
and confirmed to have the same result (being able to exceed 500).