The idea is to avoid a big spike at midnight (or whenever the local off-peak time/rate starts), which many use to set the start time.
With more sophisticated software (which currently isn't really necessary), setting a need-by time would allow the on-board computer to do something fancy, including randomizing both start and stop time. Given cooperation between Tesla and utilities, the car could even get real-time info about the charging time and rate that is best for the grid. All within the confines of making sure the charge is complete at the specified need-by time. Of course, the driver should still have a charge-ASAP option when needed.
This isn't a "surge defense" approach... continuous charging loads are still going to pile on top of one another. In network protocols, we put randomization in place to ensure that we don't get oscillation effects - but that's not the issue here. The issue is total aggregate load of significant continuous loads. Whether you start at 12:00, 12:10, 12:30, or 1 -- or if you back-load it, whether you finish at 6:30, 6:45, 7:15, or 7:30, there will be a significant peak point. My hypothesis is that you simply won't see the "smoothing effect" that you're looking for.