Garmin Engineering gives us some answers:

Let's start with the "measured LESS than the actual distance side of the issue. 

1. If you are traveling less than about 2.2 mph (3.5KPH), and you have an "SA Era" unit such as the G-12, G-48, G-II, G-III and others of similar age,  then  your GPS will likely not record the distance traveled into the log. This is because below that speed, the normal perturbations of the satellite position coupled with the reception of multipath signals would likely have added enough erroneous distance to cause a significant distance traveled OVER measurement.  A different algorithm is used in units produced AFTER the SA era,  but even in these,  you can lose some distance  if you walk very slowly.
2. Another item to consider is the Power save mode.  This can cause a shortcutting of corners that will also result in a shorter distance logged.
   

Now let's discuss the "measured MORE than the actual distance traveled" side of the question.

1. These cases are more rare. Logging LONGER distances traveled can occur due to multipath. This is especially true where a) you have buildings, mountains, trees or other objects, shadowing one or more of the satellite signals. This can result in reflected signals periodically traveling further than the direct signal and causing your GPS to think it has moved (sometimes a thousand feet or more) when actually the motion was much less. This would be much more likely in cases where the GPS had only 3 or 4 satellites locked.
2. Another possibility depending upon the Users speed is how the unit will continue to "dead reckon" when reception is lost. The unit will assume that you are continuing on the same heading at the same speed until reception is regained or 30 seconds has elapsed. At this point the unit states reception is lost.  However, the user's travel distance may not have been so far if the user slowed down during the dead reckoning interval. Still, the  distance that is "traveled during this dead reckoning" is also added tothe trip distance.