It is certainly not impossible to build accurate power meters with asymmetric crank arms, but it seems impossible to reach the claimed accuracy when manufactured at scale. When even manufacturing tolerances that would otherwise be considered small make it difficult to produce accurate power meters at scale, the fault lies with the design.
This is further evidenced by the fact that according to @gpl’s testing all Shimano-based right side power
meters suffer from the same inaccuracies.
I don’t think this is correct: the properties of carbon are highly anisotropic, so properties like stiffness depend crucially on the layup (type of carbon, direction of the weave, etc. etc.). The same can’t be said for isotropic materials like aluminum.
Moreover, if it were the material, then the left side should also be inaccurate — which from what we know it isn’t.