There was another thread where one had one light(directed) a cube with a volumetric material. The perfect example to test the ray noise.
Bumping up the value to 4 had an obvious visually better result than 2. the render time didn't increase by much. I tried up to 32 which also gave much better results than 4 but the render time was more than doubled.
I'll look for the thread later but meanwhile: increasing the LSM helps a lot to clear volumetric noise but at too high values it takes way too long. A max value of 10 should not be exceeded. (according to my tests)
Found it!7 passes each. left, LSM 2 / right, LSM 8.
