Hmm. I've not had much to complain about even with ISO 3200. I guess we all have our different expectations and standards of what high ISO is and how it should look.
This thought led me to set up my camera for a simple test:
I did an "ISO test" sequence using a tri-tone chart and my bookcase, Aperture exposure mode, f/5.6, manual focus, with a custom white balance sampled from the gray part of the tri-tone chart. I stepped from ISO 100 to ISO 25600, all raw capture. I brought the nine exposures into LR Classic and examined them. On the initial sequence, I found ISO 3200 was about the limit of what I could tolerate in terms of noise and DR degradation. However, examining the histogram and using LR's exposure slider, I found I could better center the three spikes of the tri-tone chart with +1EV exposure compensation. So I replaced the card back into the camera and did the same sequence once more, this time using EV +1 in capture. The result was a marked improvement in the noise signature such that ISO 6400 was now more acceptable and even ISO 12800/25600 much smoother and theoretically usable under extreme lighting circumstances, when much reduced DR wouldn't be a problem.
This points out to me that the meter calibration is about a stop down from what one can get out of the sensor in practical terms using raw capture ... most likely to protect the quality of JPEG output, I would imagine. A similar test done with my Panasonic GX9, Leica CL, and Olympus E-M1 show similar "protective" under-exposure metering calibration on all of them, with the Panasonic being the least protective and the Leica CL being the most.
I then looked at the images with the in-camera display and review histogram: after a couple of tries, I found the "high exposure" warning flipped on with ISO 100 at EV +1.7 and re-shot the sequence at that setting once more. Reviewing them in-camera, I saw the High Exposure warning come up on about half of them, so I shot the sequence one more time with EV +1.3 setting.
The EV +1.3 setting nets the most data at high ISO settings, with even ISO 25600 capable of returning some satisfactory results once you apply noise reduction filtering, at the cost of a bit more dynamic range (top end saturation limits, essentially). Of course, you'll never get the same dynamic range and smoothness/detailing you do at ISO 100 to ISO 200, but for some subject matter it would prove quite acceptable (at least for me).
Fun stuff. I understand better now what I can get out of the 907x sensor and how to manipulate the taking settings to optimize my results for a given subject type.
Thanks for inspiring me to look further!
G