Well yes it always astonishes that people want to mutiply in the occlusion pass in last. It somehow miraculously does indeed work (but not perfectly, it gives a very miniature, and unrealistic look to things) but gives a very subtly different look then its intended. Theres really many articles out there that say it should be done this way by the way (so much in fact that it took me most part of a month to explain it over to a comp artist before he said blimey your right, google is a wonderfull tool if ALL data indeed was reliable and right).
Like i said your organization would indicate that occlusion is some kind of shadow. Which is true in a sense. But occlusion is also simultaneously a light. Now the indirect illumination models all are problematic in a way for comp, because in their nature they convey 2 things that cant be separated form eachother.
The main problem with this approach is that you have a lightpass + shadow pass which represent shading, shadow and light together. And occpass thats shadow, ad light (its dabatable but maybe even shading). Now assuming your light pass is actually a full color one that reacts to diffuse color coefficient (which i wouldn't do if i were you). Now your occlusion pass is some sort of free thinker here and you have the possibility to over saturate it. The most problematic thing here is however how the 2 shadows interact each other.
Both are in a way grasycale values from 0 to 1. Now suppose beth have a value of 0.5 at some point it now produces a value of 0.25 However that would make the shadows darker than intended. Well as it happens this is allmost fine if you never actualy use values of above 1, and always balance the lightning off the occlusion to certain magnitude of shadows impact (however no freedom here). So in way this forces you to blend in the occlusion by boosting the occlusion shadows blackpoint into 0.5 trange for a 50% contribution or your realy overflowing Making it a dirt map insted of a occlusion light.
Furthermore because occlusion tends to increase on glacing angles of organic shapes it tends to exaggerate sharpness in the image because the edge aa is accounted by multiplication.
PS yes toby i multiply the shadow passes in (i render then directly as negatives, because thats what the shadow operation outputs no idea in inversing 2 times), and as 1 channel ONLY, unless i realy realy need colored shadows in wich case its realy hard to quantify the passes at all. And yes each light os separate for me, as is each lights shadow. (last render i did spit out 150 passes, some of them floating point some color and some masks, coverage, as well as zbuffer, normal bent normal, roughness, fisrt raylength etc... all at subsample level by the way but i dont do much rendering anymore) But basicaly i could change allmost anything in the scene afterwards.