Always prefer load data infile for importing datasets, the inconvenient is => you have to create a table structure before importing. Import wizard allows you to create on the fly a new table directly from csv or json.
I think the reason of this slowlyness is : workbench uses python for the import wizard. You can see that in the log of import wizard when an error occurs, it's python's console log.
If you don't wan't to create the structure for any reasons, you can start the process, it will create the table from the csv, then kill the process. Then, delete everything from your table and load data infile. It's a kind of ugly "hack" but it worked for me.
COMMIT
after every insert . – Cyaneous