In linearisation I have known people to clip low % to avoid boxes around stock images that are not clean white. In Quark there used to be (sorry i am not updated to say if it still is) and option to clip colour info below a certain %.
We never had problems with the Azura plates at 2% and were getting good stable dots even below that, using hybrid raster, the biggest problem getting measuring device to measure correct. We actually did the test with step patches of 0.5% from 5 - 0, our main reason for doing this was to eliminate the risk of ugly cutoffs in gradients.
That being said, we sometimes send jobs to be digitally printed on an indigo, and there we do find cut off in low %, exactly where depends on how fresh the blankets are (new blankets being to hard and giving cutting off highlights more than those that have been worn in. Best results being just before they are worn out ;p ).
That all being said a 2% with a dithered pattern or a coarser screen (lower lpi) will usually show up more consistently. It is not unusual for printers to give in to customers increase lpi beyond their equipments reliable values. Moving from 120 to 200 lpi will make the dot size 0.36, wich ofcourse means if your technology is limited to produce 2% reliably at 120 lpi you may find at 200 lpi you're hitting the limit at between 5-6% (hope i did the math right) and that's not even getting in to the lpi dpi discussion.