I think the next step is to compare capt seq hook code on this camera to other similar generation cameras that don't suffer the problem. msl reported that the sx220 does not have the same problem, although it has some other issues.
I spent some time looking at this. Aside from addresses, the canon capt_seq code for sx220 looks identical. The CHDK hook locations are also the same except the nr/darkframe and remote hooks being in a slightly different position. Those shouldn't matter for this problem, especially if those features aren't in use.
Other r47 cameras have the raw hook in a similar location too.
Regarding the sx220, msl said
The fast shooter works very well. With the default settings I get 5 JPGs and 5 DNGs with the same numbers in the file names.
In continuous mode the camera crashes after the first picture with the activated RAW/DNG option. This picture is saved as DNG but not as JPG. Then freezes the camera. A ROMLOG is not available for this crash.
It's probably more of a problem of the memory.
Freezing without a romlog could potentially be a semaphore issue.
edit:
It would be useful to know if the dark frame control and remote hook work on both the sx230 and sx220.
edit:
Thinking about this more... the fact the DNG header is sometimes saved suggests a timing issue, where something prior to the raw hook allows a task that does file IO to run, and it sometimes only gets ahold of the semaphore after the DNG header is written. The DNG code sleeps after writing the header to allow the reversing task to start up.
This in turn suggests that moving the raw hook earlier might avoid the issue, but I didn't see any really obvious places in the preceding function. There are multiple code paths between the remote hook (which if it works must be before the raw is created) and the end of the function.
One option might be to add debug code that just checks the semaphore status at various points.
It's puzzling that sx220 doesn't suffer the same issue, but if it's dependent on timing card speed and shooting settings could be a factor.