A racetrack would imply a constant bias on either side, thus a uniform distance normal to the curve. While that is quite computable, the result would not be easily representable as a spline.
If you want a simple constant vertical shift, thus a constant offset up and down, then just do three spline fits.
pp.base = pchip(x,y);
pp.upper = pchip(x,y + delta);
pp.lower = pchip(x,y - delta);
In fact, there are even simpler ways to accomplish a constant offset, by working with the spline coefficients themselves, taken from the pp form of the spline. But the above is simple and easy to work with. Since pchip is fast, unless this was a computational bottleneck, the alternative is probably not worth showing you.
Again, the constant vertical shift will NOT produce a pair of curves at a uniform normal distance above and below the curve. That is something a bit more difficult to achieve.
I have no idea what you want by "scaling" a spline. Yes, there are certainly things you could do. If you want to do some sort of proportional scaling, then you need to say what you want, if the above is not adequate.
