I encountered and reported this bug over three years ago. I decided not to write about it but considering that they still haven't fixed it...
http://chadscira.com/post/556999d91cb00914380006ee/Re-Starbu...
http://chadscira.com/post/556999d91cb00914380006ee/Re-Starbu...
You say in the end they "fixed" it (quotes included) and you yourself tested it and it did actually seem fixed... but it's not according to OP.
I do apologise if I've missed something blatantly obvious here!
Was your method the same as OP by the way?
Either way they should be paying more attention to this type of exploit, considering the history of the issue. I think they may have updated something and had a regression, the big question is how long has it been like this.
Believe it or not, there's a lot of software organizations out there that still get away with using anti-patterns like: not integrating code frequently, not using modern source control systems, not using automated tests, not doing security audits, and not doing incident reviews.
I can only assume this is because gift cards are intended to behave like cash (when the balance is zero, they have no additional value), but operate under the mechanics of credit/debit cards (asynchronous transactions, network failures/delays, etc.). Debit cards have the feature of going negative with very clear legal liability assigned to the account holder. Gift cards holders are essentially anonymous, so recourse may be more difficult.
One of their first concerns should be is any facet of this affected by race conditions? It's not that complicated of a problem really..