In the urban area in which I live, walk, run, bike, and do most of my measuring, the available paved paths are less than adequate to the heavy demand for off-road recreation space. XC and trail running, consequently, have grown rapidly here over a couple of decades. Many training runs and races in this region are conducted on unpaved single track. It seems to me that such a narrow path would eliminate the objection of "lack of repeatability" for measurement. Stream crossings can be measured with a steel tape, and the distance added to the counter number. If we measure such courses in winter, when there is no overhead canopy, on a course with good sky view, GPS coordinates to identify timing points or turns would automatically work, even if the exact point as recreated would lie a few feet off the trail. The trail itself defines the measured path.
I scratch my head about the experiences of my brethren here who report difficulties with off-road measuring. It's more work than measuring on roads, certainly. My experience leads me to believe that experienced measurers who measure off-road courses with attention to detail in recording the path and in mapping can produce "certifiable" off-road courses that are reliably re-creatable and which are never short and not excessively long.
In addition, in contrast to Mike W.'s experience, I have measured off-road courses in which I identified many course points with engineer's flags. My second measurements have come out to well within .08% every single time. I then record these points with GPs coordinates. With one XC course I measured this way, I perennially perform the course layout. My wife Judy and I create a wide course with several hundred engineer's flags on both sides of the course. The SPR is defined by and confined by the flags. Course marshals ensure participants stay within the flags during the race. I will wager that this 5K is not even an inch short. Historic finish times by known athletes support this contention.
Further, given our apparent position of long courses being acceptable, it seems we could simply create some criteria with which to "pad" the measurement to prevent a short off-road course altogether - just as we in effect do for a measurement of a road course with two-way runner traffic. If we have doubts about the commitment of the race organization to perform proper course layout, we measure the full tangents on the road. Of "course", if the route is not laid out properly, this is not our official responsibility. But we ensure that the course will not be short, even if it may turn out to be longer than we may prefer.