See Brad's blog post here:
Basically, by removing the columns you don't need you're reducing the volume of data that's being extracted.
You can find other posts regarding performance improvements in the improved yet still TOO HARD TOO FIND search engine.
Columns that are hidden before the extract is created will not be queried from your original database (it's not a select *, it's a select and then a list of field names). This will make it faster to get the data from your original extract, faster to create the extract file, and a smaller resulting extract and TWBX. If you are using a TWBX, it will speed up opening the workbook since the extract file is zipped inside the TWBX.
Fields that are in the extract but unused don't have any performance impact other than that they could be removed from the extract.