You might have meant this by "intensity", but you're going to want to vary lightness, not saturation. They should be distinguishable in grayscale.
For everyone else: there's a color model that is occasionally used in computer vision called Hue Saturation Intensity, which pretty much describes the theory of color that andrewcooke and bornon5 are talking about.
While I have never fooled around with it, I think using the LAB colorspace could be interesting as well. As I remember from my psy course, the two most common forms of colorblindness are red/green and yellow/blue.
Given that the LAB colorspace models these exact axes, it could be interesting for something like this.
But as I said, I have not yet investigated a lot in this direction.