We know how many common instructions we found, and we know how many instructions the loop code contains. So, if the number of common instructions is more than half of the instructions in the loop, we decide not to parallelize. Instead we call the sequence parallelizer to have the loop code parallelized if at all possible.
This heuristic works for now, but we will need something more sophisticated later. If some cost estimate is developed (it is also needed in the sequence parallelizer), it would be easy to ``plug in'' here too.