After discussing with Mozilla I wanna test to minimize the impact of the WebDriver when we run the tests. At the moment it works like this: We start a test and the WebDriver gives us back control after the onload event. Its unclear how chatty the communication is. But when we get back control we run a JS every 200 ms to check if everything is really finished.
Another way of doing it is to configure the driver to give back control immediately and then have a configurable wait time before check if the page has been finished loading and then have another configurable poll time. That we can minimize the impact of other things that runs against the browser at the same time as we runs the tests.