I've explained it here:
https://tinker.koraks.nl/photography/sickly-colors-the-crossover-issue/
That Phoenix has crossover is not a secret; it's pretty clear from the datasheet. Look at the characteristic curves:
View attachment 356647
I've drawn in a magenta and a blue line as reference points. Suppose you expose a grey patch of a single density so that it falls at the magenta line in the curve ('low exposure'), and then another exposure of the exact same grey patch, but this time so that it lands on the blue line ('high exposure'). You'll see that the densities will shift around, and more specifically, the
ratio between the color channels will vary. The densities of the low exposure will be [R,G,B] = [0.75, 0.85, 1.15]. For the high exposure they will be [1.95, 2.3, 2.4]. The base (where the curve crosses the vertical axis) is at around [0.5,0.5,0.9]. If you subtract the base (let's assume this is constant) from the example exposures, you get net densities for the low exposure patch of [0.25, 0.35, 0.25] and for the high exposure [1.45, 1.8, 1.5]. If you then take the ratio of blue to red (B:R) and green to red (G:R), you'll see the difference, as this would work out for the low exposure as [B:R, G:R] = [1.40, 1.00] and for the high exposure as [1.24, 1.03]. As you can tell, it's especially the blue channel that is shifting around, which of course shows up in the plot as well.
From here, it's a small step to imagining what happens if you photograph a greyscale that occupies not a single spot, but a range of exposures in the H/D curve. You'll see that the dark patches of the greyscale will render a different color than the lighter ones. Put differently: if you were to color balance the exposure (in a scan or print) for a neutral gray in the low values, you'd end up with a color cast in the high values, and vice versa. This also shows up if you scan several frames with different exposures at once or with the same settings and then apply the same inversion and color balancing curves to them. The colors will shift around depending on exposure. This is what you see happening in the video, although the explanation seems more intuitive/subjective while the one I give above is a bit more technical/(quasi-)quantitative. It boils down to the same thing, though.
Coincidentally I'm currently working on a blog that illustrates this for color RA4 paper which crosses over since it has become optimized for digital. However, the crossover is far, far more subtle than in this film and it's only noticeable if you look for it. In Phoenix film, it's actually challenging to color balance even a scanned image to avoid this crossover and in printing, it's virtually impossible even if you were to mask it, since the crossover is also non-linear.
It's technically speaking an atrocious film, but creatively speaking, it's a unique proposition.