Previously transforming and translating meshes (used by the object info and transform geometry nodes) was single threaded. Now use the same code path as other geometry types which already includes multithreading. I observed a 5x performance improvement for a 4 million vert mesh on a Ryzen 7950x.