Ah yes, that last one from Luc is something I've also found to be important: "In density fractal/tweak noise : reduce noise variation (<0)". Noise variation can actually be a hindrance when trying to make regular looking billowy cumulus. Maybe Luc suggested trying negative values, I am not sure, but there might be some benefit to that too.
Recently when I've tried to make cumulus clouds I've used the following general approach:
Cloud layer "Height modulator" "Depth modulator": A fractal with noise type "Perlin Billows". Noise variation = 0. Roughness between 1 and 1.3. Possibly increase the Coverage adjust between 0 and 0.75.
Cloud layer "Density shader": A fractal with noise type "Perlin" (or sometimes "Perlin Ridges"). Some noise variation is OK, but not much. 0 is probably safest. Roughness between 1 and 1.3. Possibly increase the Coverage adjust between 0 and 0.75.
The low roughness of this setup means the edges won't be very wispy, but the interplay between the Height modulator and the Density shader makes some interesting variation.
The edges can be made a little wispy by putting another fractal after the first Density Shader, with a higher roughness and very low contrast (less than 1/10 of the contrast of the other fractals). If you're far away from the clouds, roughness of 2 might be good. If you get really close into the clouds, an extra layer of higher roughness (e.g. 3) with very, very low contrast (say 1/100) can be nice.
Another method to add wispiness is to use the "Final density modulator" input with a Coverage adjust of 1 (so the values modulate above and below 1), high roughness as above, low contrast.
I'll see if I can put together an example.
Matt
