There shouldn’t be any drawback, the vectored functions were designed especially to make it possible to split headers & payload, even across layers. It’s considered good practice to use them.
We didn’t actually measure impact, just assumed that juggling larger buffers and having to copy data more often is more expensive on our little MCUs.