We don’t see any evidence in favour of either scenario at the moment. The biggest point of failure we see is the hardware connection, although not due to its design. It’s mainly an issue of heat, wear and tear with media cards and connectors. When verification errors show up in a transfer, it gets redone, and most of the time either the error vanishes or it pops up in another spot. The latter is a sign of an unreliable medium or connection. If it pops up in the same spot, which rarely happens, it’s a sign of bad sectors.
Theoretically, the chance of unrepairable failure is spread out much better with DNGs than with larger files: if you have a corrupt file, it’s just a frame that’s easy to drop. On the other hand, doing a lot of I/O reads is taxing, so it’s a trade-off.
In our opinion, there’s is no safe file or format or even storage medium. It’s just a matter of keeping in mind that data is volatile, and definitely not set into stone.