Problems when solving weights for geometries with multiple shells

[combi]

Hello,

Has anyone any trouble with geometries composed of multiple shells?
We are using it to solve weights for a given mesh sequence and bones animation,
like in the following example in the data folder:
# Solve skinning weights from input meshes sequence and input bone transformations
$EXECUTION_FILE -i="Bone_Trans.fbx" -a="Bone_Anim.abc" --nTransIters=0 -o="SolvedWeights.fbx"

When using a geometry with multiple shells, under certain (unclear) conditions, the solve quality is bad.
If we rework the geometry to have only one shell, by connecting the shells together, the same command will produce a much better result.

I tried many of the settings, but it seems many of them have no effect in the mentionned context.

If anyone has any pointer about dealing with geometries with multiple shells, it would be great.
In the attached archive, are fbx/abc files, and the shell script used to show the problem.

Upon inspecting the "geo_two_shells_bake_result_wrong_solve.fbx" file you'll notice that some parts weights are really off.

Of course reach out if some more information is needed...

Nicolas
multiple_shells_problem.zip