You are not paranoid. It's unclear to me how SA_MAGIC can become corrupt in the first place.
My working theory is that a bad byte in memory or a failing controller wrote the wrong header to disk in a redundant fashion. The data block then ends up looking consistent in checksum/redundancy, but is an invalid ZFS block and causes the kernel panic when it is read out during a file access.
I don't think my Oracle contract covers the piece of hardware in question and therefore I can not submit an SR. I don't even see bug #15850819 to tell you my opinion of whether it matches.
I'm going to turn NFS back on tomorrow to see if the resilvering/scrubbing managed to fix the bad block after all. And, in the meantime, ensure full backups of everything I can successfully read if I have to rebuild the zpool.
CR 15850819 is private and so its content isn't visible from customer MOS portal at least at the moment .
But it couldn't stop you from rising an SR against CR 15850819 and obtaining the fix.
Feel free to submit SR 'cause root cause is software problem, not hardware.