I update now to latest 1.6.0. Also, no problem with 12s.
Which Mode do you use? Flash?
Hmm.. strange..
I did the last view test in Av Mode (set in CHDKPTP, the Camera was in M Mode). I use AV mode for sunset and M mode for sunrise.
EDIT:
I changed the ui_raw_hook_sleep=9000 and now I was able to shoot up to exactly 20 without shut down and with 21 the camera again went off.
It looks like waiting ~10s in the remote hook is the problem. The script calculates how long it needs to wait based on how much time all the work (shooting, processing the raw data, saving the file) took, so you probably want to leave some margin for variation.
But, this seems to confirm moving long waits to the raw hook probably addresses the problem, so the script could be updated to allow arbitrary lengths, as long as there's not some similar problem in the raw hook. I think I actually considered this early on, because blocking for a long time in the remote hook is quite sketchy, but didn't bother because it seemed to work.
You should be able allow arbitrary ui_raw_hook_sleep by changing the line (1991 in rawopint 0.25)
hook_raw.set(10000)
to
hook_raw.set(10000 + ui_raw_hook_sleep)
That said, rawopint metering will degrade with long intervals, since it bases exposure on the previous shot. For this kind of interval, it might be better to use camera AE, or take (and ideally, discard) a metering shot immediately before the real shot.
The first two romlogs refer to CntFlashTask0, which I don't recall seeing before. The name suggests it's either related to flash (strobe) mode, as c_joerg suggested, or possibly flash memory.
IIRC the MotionVector.c one has been seen before in connection with the remote hook. A quick search found
https://chdk.setepontos.com/index.php?topic=13893.msg141682#msg141682 but I'll have to do some digging to find the details.