It can be useful to increase default gap penalty slightly. Try values of 9 or 10. The length dependent statistics are reasonably similar for these values.
You can use the simple method to scan, then find out what the score threshold would have been for a NALL scan (by using the -F1 option to scanps). You can then extract all sequences that give scores above this threshold and run the NALL method on them. This two step strategy is usually quicker than running the NALL method on the whole database.