The reason you didn't find any articles and haven't gotten any responses on RG is perhaps the ambiguity or vagueness of the question. FEM (for example triangular or tetrahedral elements) is often used with objects or domains having curved surfaces. Representation of the domain using this type of element inherently accounts for the curvature of any surface, which is not the case with FDM. Surface curvature is therefore not an issue (nor is model stability), which is why researchers might use the FEM in the first place. There are many articles comparing FEM and FDM for problems with curvature, including this one, which I published 30 years ago comparing laboratory data and two CFD methods (staggered and boundary-fitted elements).
I came across this paper "Dealiasing techniques for high-order spectral element methods on regular and irregular grids" today and I remembered this post.