[dgaies]

Thanks! That's conclusive info that will help the OP much more, I'm sure. Do you think the explanation is because of the buffer model? (see buffer explanations here)

Chances are that from ISO6400 and upward, some heavy(ier) NR is applied on the data in the buffer, before it is sent on its way to the memory card. If the K-5's buffer logic is setup this way:

that would explain the slowdown. TANSTAAFL, I suppose. The laws of physics are not bendable, not even for Pentax

There's been some discussion of this over at the "other" forum

I didn't get too far into the technical discussion, but I believe that you are correct and at 6400 and above there is additional NR being applied (not sure if the NR applied just to the embedded JPG for display on the LCD or to the RAW data as well). So I would guess that for this additional NR to take place, a copy of both the unprocessed as well as the processed image needed to be on the buffer, which then might explain why the reduction is exactly half at 6400 and above.

Also, note that the drop from 8 to 4 takes place with all NR settings turned off in the menu as well. Figured that would be the next question

[kittykat46]

.....
Chances are that from ISO6400 and upward, some heavy(ier) NR is applied on the data in the buffer, before it is sent on its way to the memory card. If the K-5's buffer logic is setup this way:

that would explain the slowdown. TANSTAAFL, I suppose. The laws of physics are not bendable, not even for Pentax

There may well be NR applied at > ISO 6400, but I can see no evidence (so far) of heavy-handed NR.

K-5 images at ISO6400 and even 12800 still show a high degree of detail.
Remember, TANSTAAFL applies to NR as well - cleaning up the image with NR will involve loss of detail.
Unless Pentax has developed some very good algorithms for in-camera NR with minimal loss of detail. I'm more inclined to believe there is not a lot of forced NR going on.

[sterretje]

...I think that's a reasonable assumption based on the prices of flash memory these days

It's not flash; would be far too slow. It's more than likely dual port SRAM.

I would see that proven please - seems highly unlikely, ridiculous actually as 1GB equals a whopping 100-125 jpegs rather than the 30 quoted.
...
...
Your math is way off because while the buffer is filling up from the sensor side, it is at the same time emptying to the camera's SHDC drive.

I agree and guess that 256MB is more accurate (roughly 30x jpeg).

[Pentax Lover]

That is because It's a strange question. The camera is a computer, is it software related?
Naturally there is hardware limits and of course there is software running that uses the hardware.

Thanks for clearing that up for me.