Everyone sane is using the smartcard interface because usually the chips cannot be cloned whereas a magstripe writer sets you back $150 and about 10 cents for a blank card. Just for the lulz: many payment terminals allow you for reading the WHOLE RAW STRIPE DATA by a command... so you hack the PoS software to issue a CardSwipe command before issuing the payment command. Unsuspecting store clerk will follow the instructions on the terminal: 1) swipe the card, 2) see the terminal saying again "Please swipe card", 3) think the first swipe failed 4) swipe again and do the normal payment. This is trivially done with any ZVT or OPI-based payment terminal. And it fucking works. (If anyone is interested, drop me an email. I develop PoS software and can send you a demo video)
The smartcard interface is the thing where the problems lie: PC/SC is a quite easy API, but most of the software will be written in C and suffer from the usual bugs: boundary checks done wrong, overflows, missing sanity checks... this uber-long PCB is essentially not needed at all. You could entirely implement everything on e.g. a BasicCard ($10 apiece for the high-memory+RFID card) or any other programmable smart card.
Edit: I've been asked by email by some people for the demo video. It'll be ready by Tuesday, I'll also submit it to HackerNews.
A commenter on the Krebs site make the same point about the smart card interface:
http://krebsonsecurity.com/2014/05/thieves-planted-malware-t...