That's another great display that TI uses silly, slow algorithms...
That reminds me of Samuel Stearley's faster next_expression_index() and list item access routines for the TI-68k series.
Out of curiosity:
* how much execution time does inlining the comparison function save ?
* what's the execution time of an implementation of shellsort with a reasonable sequence (not the original O(n^2) one) on the same-sized lists produced from the same random seeds ? There's no O(n*log(n)) implementation of shellsort, but there's usually a range of sizes where an implementation with a reasonable sequence it trounces the quadratic sorts while being competitive with some asymptotically better O(n*log(n)) sorts. 200 items could be within that range, but 999 usually isn't.