I’m sorry, there’s no real news; upstream ticket. I haven’t managed to find the root cause, and apparently it hasn’t happened on my Omnia during the past three weeks, so it’s more difficult to uncover without a reproducer.
I believe it would significantly help me if I had a cache file from a moment when a name fails this way, because by spoofing system date I could repeat that failure at will and inspect what exactly is happening. On Omnia it’s located in /tmp/kresd/*.mdb.
Sending the cache is some privacy exposure, as it contains names that you queried and some timestamps on them, etc. – though I’d promise to only look at the failing name and names of nameservers required to resolve that name (transitively).