Rough estimate of the distance from a plane to the last molecular collision

Distance until next collision is exponentially distributed, as is distance since last collision. The mean value of this distribution is $lambda$. The direction from the last collision until the point at which the particle hits the plane is assumed to be distributed uniformly over the sphere. Together, this characterizes the full distribution of the location of the last collision, relative to the current location of the molecule.

Let the plane of interest be the $xy$ plane, and let the particle hit it at the origin. Then the location of the last collision is distributed with probability density proportional to $e^{-frac{3sqrt{x^2 + y^2 + z^2}}{lambda}}$. The constant of proportionality is given by $$ frac{1}{int e^{-frac{3sqrt{x^2 + y^2 + z^2}}{lambda}}} $$ since the distribution must be normalized. The distance from the plane is just $|z|$. The mean distance from the location of the last collision to the plane is then $$frac{int |z| e^{-frac{3sqrt{x^2 + y^2 + z^2}}{lambda}}}{int e^{-frac{3sqrt{x^2 + y^2 + z^2}}{lambda}}} $$

If you want to try these integrals yourself, change to spherical coordinates and integrate the radial terms by parts. I find $frac{1}{2}lambda$, so it seems likely that something is wrong above.

Edit: The book you reference cites chapters II and III of this https://archive.org/details/in.ernet.dli.2015.1789/page/n151/mode/2up for a derivation. This book gives the $frac{lambda}{2}$ answer, and then proceeds to explain why it is wrong. It turns out that what we are interested is not the distance between a plane and the last collision for a molecule chosen uniformly at random. Instead, it refers to that distance for a random molecule chosen from all molecules that cross the plane over some short time. Some molecules are more likely to cross the plane than others, so this is not a (uniformly) random sample.