This is because /opt is on a different Qubes OS storage volume (the ‘root’ volume), which in case of an AppVM comes straight from its TemplateVM.
If you limit the VM selection for restore to just e.g. the “wallets” VM and try that let’s say 5 times, does it always fail at exactly the same e.g. .009.enc chunk? Then that chunk is likely to be corrupted. You might still be able to pull out enough data from the other uncorrupted chunks if you’re lucky:
However if the error sometimes occurs at a different chunk, then the data itself might be fine and it could be due to something like overheating, bad RAM, or maybe a variation on this recent incompatiblity with some AMD systems.