After a rework of the precomputed Lagrange points by @jsign the performance has improved, at the cost of a larger file.
