I've been doing some research shots with S3's raw. At the first time I was using Drifft to convert all raw files into (unfortunately only) .tiff, since it cannot produce the .dng format (though it also made dng and jpg folders instead of tiff). After couple shots, I still couldn't find a consistent conversion results since drifft wasn't previewing some settings. I was only getting one very nice iso400, low light, indoor shot result compared to the jpg's, with a significant difference in noise grain (all the raw are almost always have finer, relatively smaller grains of noise). The other shots were not promising at all. I was about to give up, but eventually found the "DNG for Powershot 1.1.4". It was very relieving to get the program works without any problems. Working with .dng, bring it into photoshop's ACR, the adjustments sliders are abundant and also previewable.
The whole new world begins. The raws are just what they are promised to be. I took extreme test shots outdoor under harsh midday sunlight, and low-light indoor. I managed to recover things more than I expected. In photoshop's dialogues, all settings are left unadjusted but the "sharpness", "recover", "contrast", and "noise reduction" are properly adjusted (except for contrast is almost always set to minimum, and sometimes the recover is set to maximum, and maybe due to default ACR's noise reduction the noise is significantly lower compared to Drifft's. But if you're lucky enough to own more advanced programs like neat image etc., I'd recommend you to leave 100% noise reduction to them).
Unlike once said that small sensors are just useless for its raw, it surprised me that those sayings are not completely true. Completely white surfaces reveal their textures and colors, unsaturated shadows reveal their true colors, lightings are more balanced, and noise characteristics are also a lot friendlier to manipulate. However, some cons are also emerging. The appearance of chromatic aberration in some unexpected places, possible occasional aliasing in some edges, possible highlight banding due to "over-recovery", hot-pixels in long exposures, and sometimes a lot softer images compared to the original jpg. The unsharp mask with 100% strength and 2.0 pixel radius is closing the gap, but not as good. I'm still researching on that.
Watch closely to my sample image attachments. 4 crops of each jpg and raw, side by side.
The first 2 are high contrast outdoor shot, the next 2 are low-light long exposure indoor (15 sec).
If I may summarize the characteristics of S3's original jpgs:
1. No matter how you set the in-camera-contrast adjustment, the whole images are always processed relatively in high contrast (that's why I won't set the contrast higher than 2 - it will hide things even more in hilites and shadows), results are lower dynamic range.
2. Low-light images or even most deep shadows in bright images are highly desaturated, losing their original colors that you can't recover no matter what (btw, the iso800 itself desaturated around 25%). Try boosting up the brightness or gamma and also the saturation of my jpg versions, or shoot and find it yourself.
Have a nice try yourself! Sorry for the long post...