Note that the CRT points on the CIE chart are not pure colors (on the very edge of the color area) but have mixture of different colors.
Looking at this more precisely, the alignment is not with the peaks of the each receptor separately, but with the wavelengths where each receptor is most nearly the only one excited. In the plot shown on this page, these are the peaks you get if you divide each curve by the sum of all three curves. I found the data which generates those curves, made a fifth column which is the sum of the 2nd, 3rd, and 4th columns, then divided these columns by the fifth column. When I plotted this up in Matlab, the peaks of the relative excitation curves are at 435nm (blue receptor), 515nm, (green), and 700 nm (red). From The C.I.E. Chromaticity Diagram, these wavelengths correspond to about the largest triangle you can make using three colors.
It’s also easy to see from this why some colors can’t be made from these three colors: at the green and blue peaks (435 and 515), red is being excited, but in between (around 500 nm), red goes through zero. This is the region left of the line between 435 nm and 515 nm. Likewise, the blue receptor is excited somewhat at 515 nm, so you can’t make a color corresponding to, say, 560 nm (where the blue receptor is essentially not being excited) from 515nm and 700nm, because you’ve got some blue that shouldn’t be there