1) might be the same drift issue others are reporting, complicated a bit by routing with lock on road. If it really is the same issue, presumably "lock on road" was responsible for correcting the issue without having to power-cycle the unit. Or it might not be the same issue at all.
2) Not sure this is the same issue. This one really sounds like plain old high-sensitivity "wandering", although the distances are a little bit large. Also, unless Garmin actually FIXED something in 2.6, the drift issue never clears itself up without power-cycling the unit.
3) Not sure I understand this one. Failure of the return track to match the outbound track would be characteristic of the issue others see. Not sure what you mean by the reference to "another trail, which seemed to be not on the map", though. And 100m is rather short by comparison with the distances others report. But anything is possible. If this is the same problem others see, no one else has reported it as correcting itself without a power cycle.
If this IS the same issue, it has been there for a long time prior to the 2.6 unit firmware. There is some evidence that it was introduced with the 2.60 chipset firmware, which has been around for quite a while. Although it appears that some people see the issue repeatedly (frequently in the same location), it is rare on the average. So it's possible you've had the problem all along and simply never tickled it before.
Thanks for your thoughts, here is a little bit more...
1. I always have "lock on roads" turned off, since I only have a non-routable topo map. So it wasn't this feature, that was "healing" the signal.
2. I will try to power cycle next time if I will see an effect like this again.
3. I have a screenshot for this one, but I don't know how to post it here without having webspace... The 100m of back and force are absolutely on the same track and also on the path painted on the map. Then we turned to the right leaving this path, onto a path that was surprisingly not on my map. During our little break, the triangle suddenly jumped onto a path painted on the map. The shape of the track looked exaktely like the path painted on the map, but it was shifted about 50m to the west from the time we turned around. You will understand, when you see the picture. It is not map accuracy, because the rest of the tracking was right on!
I might be wrong but I still have the feeling that the drift was not there forever. Could it be that the hardware is getting older and is running out of some internal calibrations and tweaks? Some bathtub curve effect maybe? Just a thought...