As I understand, the CIExy graphic maps "greenness", or rather middle-wavelengthness, to the Y axis and "redness", or rather long-wavelengtsness, to the X axis. The trapping used to reduce the 3d gamut of the human eye to 2d is that the gamut is specified for a certain constant luminosity, and long-wavelength + middle-wavelength + short-wavelength equals the said luminosity, thus the third coordinate can be extrapolated from the first two.
An example of this plot can be seen here, on wikipedia.
The curved edge of the human vision gamut gives the single wavelength spectrum. If this is the case, how come the RGB gamut (triangular) does not intersect the curved edge of the human vision gamut? The light coming off the 3 LED categories in the display should be of a nigh-single wavelength (as LEDs have a single, very sharp emission peak).