They are planets - a lot of them might be moving around at once, and you might have several simulations going at once. Every planet needs a separate set of table of sines and cosines as functions of time for each instance. You don't want to recalculate these tables constantly, rather caclulate them for two times that are "close" with respect to the approximation and then linearly interpolate them, which is much faster than doing 31*19 sines and cosines for every planet and then slogging through a perturbation table with hundreds of coefficients.

So Todd's idea was the best one, you create a pool of tables and then leave them lying around until memory needs to be reclaimed for whatever reason. This is a middle ground between just hogging a buttload of memory with static tables vs. mallocing and freeing them over and over again.

The perturbation tables come from a spectral analysis of the output of JPL's DE404 numerical integration of the Solar System - so they are extremely accurate over a specific interval. 1/4 to 1/2 second of arc in 4500 years is realistic, even for Pluto. (It's an amazing thought that the exact details of the motion of the Earth, Moon and planets over a 4500 year interval can be modeled in a few hundred K of C code.)