But trying to justify such an estimate [10% solar system ejection rate?] mathematically or computationally might be very difficult, since the ejection process could take 100M years.

--To put it another way, it is somewhat like the Turing halting problem. If your computer simulates some orbit for a few Myears, how does it know whether body will be ejected? If ejected, it knows. If not, then computer still does not know whether will be ejected later. Perhaps it can prove a theorem some orbit is stable? A paper by me of a related topic is Church's Thesis Meets the N-body Problem #21 here http://rangevoting.org/WarrenSmithPages/homepage/works.html in which I showed more strongly that Newton's laws of motion+gravity for point particles are UNSIMULABLE in the sense that if you have a Turing machine and it has input tapes giving the initial velocities, positions, masses of N bodies to infinite # decimal places, then if you ask machine "will body 5 enter the following spherical region xxxx in the next hour, or will it never enter the concentric 2X larger sphere even in 2 hours (I promise it will do one of these 2 things and there will never be a collision)?" then it is IMPOSSIBLE for any algorithm to answer question no matter how much simulation time. Furthermore, the trajectories of the bodies in the next hour can have "infinite descriptive complexity" and resemble/encode an arbitrary Turing machine infinite computation. This was part of a research program by me if trying to assess computational complexity of different physics theories; I succeeded in the case of Newton laws. Amazingly, quantum mechanics was found in a sense to be EASIER to simulate than classical mechanics... see paper #49, also a success... and for a partial success see paper #58 on hydrodynamics. The simulability of more advanced theories such as general relativity remains completely open.