Dear Laurent,
thank you for your valueable post. The solution of populating the pipe surfaces with points and then using them as input points to place voxels in Dendro is working very well and stable.
What was your experience on the computational performance (e.g. required time) of the dendro approach?
In general, to design such organic tube-like objects, I see three possible workflows:
Workflow A): Populate pipe surfaces with points > Place voxels at points with Dendro > (Smooth voxels) > Extract mesh
Workflow B): Convert pipe surfaces to mesh representation > Apply thickening operation on surfaces meshes > Unite meshes as a single mesh
Workflow C): Create thickened, closed brep bodies from the pipe surface > Apply boolean unite operations on brep bodies > Unified brep body
I guess, in the end it depends on what type of representation a designer wants to have (e.g. a brep keeps the eaxct representation and allows creating meshes with any resolution). However, if output as a mesh is OK and the result is smoothed anyway, workflow A) seems to be most stable and fastest - because the unite operations in B) and C) seem to fail some of the time if no clear intersections are present and require some time to compute.
I would be curious to hear your and other thoughts … 
I tried to implement approach B) and C) with custom scripting. Corals made out of breps look like as in the attached picture. It requires some thought on how to deal with the tolerances and intersections but Rhino seems to be very stable in peforming complex booleans with many brep bodies. My experience is that breps are more stable compared to mesh operations as you have a clean intersection curve. The speed of brep operations seems okay so far (a bit slower than dendro).
