No worries. I'm trying to convey a complex nuanced mess in a non-native language. I try to expect this with patience.
As far as I can see ZSERVER_PORT is used in constructing this before it gets set, so minimally I only have to set that (or the host) to have uniques per batch. I've opted to use ports rather than 127.0.0.0/8 as the former is better supported by CI tools (and also works trivially as an approach across a wider range of systems).
The tests seem to explicitly close the browser at the end of each test.
And as for changing the testing infrastructure in regards to Selenium, the feedback and experience you share is most welcome, thank you. It'll not be a short road to change any of that. For now I'm tackling the lowest hanging fruit (from my point of view). I will get back to this later in the year.
I'll try to get this way of running the tests into a mergeable state just before or during Sauna Sprint in July.
Before that I'll try to figure out how to get the alltests wrapper parallelization gain for current DX runs. Likewise someone better versed in the art of Selenium can tackle the Robot tests. These alone could take 20min to 30min off the test runtimes.