I don't think a cloud shape is as simple as that.
You're doing 32 samples sunward for the whole sky - I don't know your true spatial resolution (I'd be interested), but it's not unrealistic to take clouds 30 km away in low light to cast a shade, so you'd have a km-sized resolution scale.
At this resolution, a cloud shape is very simple.
You need to have some kind of noise function to modulate the density in the height axis.
At a spatial resolution of 50 m which the texture provides, I think I can basically just store the value - the resolution is an overkill for shading already.
The current technique runs really fast on a low sample count, and I don't think you'll be able to further decrease the amount of work done per sample, not without compromising how good the clouds look.
Unless I misunderstood what you're doing, I think you can - you need resolution for how clouds look, but not for how they shade. The latter we can solve with no noise, just a single texture lookup per sample I bet - so in fact you'll probably be able to increase the number of shade samples.