If we are assuming the underlying expansion is as efficient then the cost comes either in the boilerplate or mistakes made when parallelising. In the case of the boilerplate the solution would be embedding some more of the linked code and reducing IPC/other unnecessary bottlenecks. However I suspect the error is in the parallelisation, which is not surprising, especially as the authors likely optimised and tested for large, not small, files.
As it is, removing the GUI is perhaps the worst thing you could do to the user. Closely followed by inducing epilepsy with that ridiculous expanding dialog.
As it is, removing the GUI is perhaps the worst thing you could do to the user. Closely followed by inducing epilepsy with that ridiculous expanding dialog.