I remember the early digital cameras that had anything resembling useful quality (megapixel) stored JPEG files that were noticeably lossy. As memory cards became more capable, this was addressed and manufacturers soon got the compression tuned just right to match the actual useful information from the sensor. After all, whose to say this sample should have been 145 rather than 146? The sampling is a little bit iffy anyway, and the pixels may line up a few nanometers to one side or the other on the scene, right? And a little softening is fine, as the optics are imperfect anyway and a sudden pixel of a different color must be noise anyway. So the JPEG was technically lossy but part of the overall capture pathway and was good enough to capture the meaningful information.
Now, we have 14-bit sensors and much greater depth and color than a normal JPEG file can represent. There is still some inherent softness, whether from DOF or diffraction, since we still have more pixels than image resolving power.
A comment someone made about shooting JPEG to save space got me thinking about how the situation has changed. Saving JPEG is shooting yourself in the foot now, because it is so much less than what the camera captures.
Is there a file format that is lossy but can be tuned to match the inherent information content of the image, that will offer a substantial space savings over RAW files? We can assume that the PC can perform significant processing on the image (not done in-camera) and do things like noise removal first. "soft" (e.g. out of focus background) should naturally give higher compression, without having to pre-identify regions of interest for different parameters. But given that Photoshop can automatically select "in focus" content, it's certainly not out of the question for the processing program to figure out how sharp each region is: that would be good for noise reduction algorithms too, for frequency separation and rejecting noise of higher frequency than expected content.
For burst shots, you could do inter-frame comparisons like video codecs can do. I think there is a new still format for the web based on a single video frame, right? Well, why not a bundle of 3 or 5 nearly identical frames?
JPEG 2000 never took off, and JPEG has more in its arsenal than the JFIF profile we use for common JPEG files. Then there's fancy stuff like Pixar's files.
Is there anything that I can teach people who find RAW files to be "too large" and keep only jpeg? What would be significanly smaller than RAW but (usefully) better than nieve JPEG?