Correct - electrically, both the CT100 and PAL receivers use a straight NTSC electrical matrix, and compensate the phosphor colors in the camera. So, rec 709 encoded as NTSC and decoded as NTSC would give the proper signals for a display with rec 709 phosphors.

Also correct, if the UHDTV has wider gamut phosphors and therefore doesn't internally limit the color gamut.

Here are some calculated results for a modern color test chart, based on actual measurements of RCA cameras.

In each square, the top 1/3 is what a standard NTSC receiver would produce when fed the signal from that camera. The middle 1/3 is the ideal perfect reproduction. The bottom 1/3 is what could have been achieved by employing a linear matrix in that particular camera.

The first example is the status in October 1949. This camera had an incorrect green response for NTSC phosphors. The green response peaked at too short a wavelength, so to this camera, greens looked yellowish, yellows looked orangey, and oranges looked reddish.
http://farm4.staticflickr.com/3833/1...e64fe7ed_b.jpg
10_49 N by old_tv_nut, on Flickr

The next result is from November 1949 - the green response was corrected to be closer to ideal.
http://farm8.staticflickr.com/7335/1...73089497_b.jpg
11_49 N by old_tv_nut, on Flickr

Last is the best camera measured in March 1953:
http://farm8.staticflickr.com/7312/1...8163f767_b.jpg
3_53 #3 N by old_tv_nut, on Flickr

When you say:
Does "could have" mean "if it were not necessary to correct for the non-unity gamma response of a CRT"? Do you think future standards based on digital cameras and displays will be able to use a linear matrix (gamma = 1?)? If so, would that make it easier to convert (using a linear transformation) between the different color spaces used so far?

No. gamma correction is always needed somewhere before the signal goes to the CRT. "Could have" refers to the difficulty of adding linear matrixing to the already complex tube cameras at the time, keeping the circuits stable, etc.

The image orthicons and later Plumbicons were linear (Vidicons were not). Proper correction of the color response can only be done with these linear signals (unless you start talking about a three-dimensional non-linear look-up table, which is essentially what is done in profiling computer printers - an impossible task with tubes). Modern solid state pickups are also linear.

LCD displays have a very non-linear response that is not related to the CRT power curve. Therefore, LCD displays are corrected to match the traditional CRT curve. Plasma displays and micro-mirror devices are essentially linear because they work by duty cycle modulation, and have to be corrected also to match the CRT curve.

Now, your first thought might be to make the system linear from start to finish, but this, due to the non-linear response of the eye, makes the quantizing steps (or any other noise that is introduced) much more visible in the dark parts of the image. So, some form of non-linearity will always be preferred. The CRT just happened to do that inherently.

Particular non-linear functions (especially for digital cinema film replacement) are being proposed to handle the widest possible dynamic range without requiring too many bits. These are generally somewhat logarithmic in at least part of their range. What is best depends to a degree on the use: is it for carrying the best dynamic range in the final projected image? Is it for a digital negative that may have a wide range of exposure and color temperature adjustment made before the final image is produced?

The ultimate system would have such a large dynamic range that you could record the absolute value of anything from starlight to staring at the sun without adjusting levels, but we aren't there yet.

Edit: Computer programs that convert color spaces do the linearization first (or do the equivalent via look-up tables and interpolation) and then do the required gamma correction. A TV set able to handle multiple color spaces will do the same, but the final correction may be for the particular display rather than a CRT gamma. The conversion between gamma corrected signals and linear and back again would not have been contemplated in the early TV cameras, but stable linear matrixing via transistor analog circuits was possible and normally done once Plumbicons came along. Undoing gamma correction, matrixing and redoing it was still too complex to go into analog TV receivers for conversion to the new phosphor colors. Once processing went all digital, we got the tremendous advantage of perfect mathematical stability for the transformations, so it is now possible to make sets that convert color space and have them all perform to spec, without the color drifting all over. This was needed in some of the early plasma sets, where the red phosphor was noticeably orangey. It will be needed again if wide-gamut sets are developed, as they will have to convert color space when receiving older rec 709 sources.

Thank you for the clear explanation of the complicated subject!!

Dave