Guide to Digital Preservation from NIST
Normally punch cards are punched with machines that actually cut the chad right out of the card and push it into a waste bend. You don't get hanging chad with that. You might drop the card deck, lose cards, etc., but as long as you keep them in order, and don't bend, burn, or get them wet this kind will read back properly every time.
That's if you keep the card reader working right - at the college computer center where I worked in 1972, a repairman had to come in and basically rebuild the card reader once a week, or it would start chewing up the cards and spitting the mangled pieces out on the floor. I'm guessing that was a few hundred thousand cards read between repair calls. At 80 bytes per card, it was a fair amount of data in those days, but not much by modern standards. This was an NCR card reader. I think IBM makes much better ones - but you'd still need a lot of punch and reader maintenance and a truckload of cards to back up a 10Gig hard drive. If you really want to use punched media for long term data storage, paper tape might be a better bet. It's more compact, it stays in order by itself, and the machines are simpler and therefore more reliable. And you can use mylar tape instead of paper, if it will last longer.
The other kind of punchcard, used in election systems, is pre-punched to leave each chad hanging by 4 threads of paper. The voters are supposed to push the chad out with a hand-tool. In 1972, the local electric company (or something like that) was using cards like this that the meter readers punched by hand as they read the meters. That computer center had a contract to process this data, but we hated those cards. Even though the users were trained (unlike voters) in how to be sure the chad was punched clean out and not left stuck to the card, you got little bits of paper fiber coming loose and clogging up the machinery. You also got chads that hadn't been punched breaking loose or swinging sometimes, so if you ran the deck through again it would read a few more holes, with maybe one or two of the old holes covered up now.
So there were three issues in Florida. One was that quite simply this was a system with an acknowledged read-error rate around 2% even under the best circumstances. In a recount you'd get a different count every time you ran the cards through again. This had been known for decades, but no one cared until they got an election so close that it mattered.
Second, voters were not trained in using the hand-punch sticks, and the flexible backing that is supposed to support the card while you punch it may have been worn-out or misaligned in some cases so they couldn't get a clean punch on the first try. Someone who understood the system would have checked the backside of the card and pulled off any hanging chad, but with a bunch of octogenarians that have never even programmed their VCR...
Third, Palm Beach in particular had a badly designed ballot. They should have known this, because in 1996 a similar butterfly ballot apparently cost Dole 19,000 votes. It wasn't enough to change the results of that election, and the Dems that run Palm Beach didn't learn anything from it because only Repubs were hurt. (I generally love it when the Demoncrats shoot themselves in the foot, but not when they make a joke out of the most fundamental underpinnings of our republic...)
See Ask Tog's article on this.