I forgot a quite practical variant:
In Blender, create a new image of the (planned) size of your texture. In the creation dialogue, select "Generated Type" "UV-Grid" instead of "Blank". That produces a checker board with fine coloured crosses on it (1 px line width). Swapping your texture out for this grid permits three things:
a) You see how these fine crosses come out. Totally blurred - needs bigger texture; super-crisp - texture size can be reduced, maybe.
b) If the squares show much bigger on some parts of your model than on others, your UV map is quite unfair against those places with the large squares. If you have good reason for it, then keep it. If it surprises you, then try to get more space on the map for these parts.
c) If the squares are very stretched, it might be an idea to change that. How much is tolerable depends on the texture at that place.
The ideal is to see squares everywhere with the coloured crosses on them just visible. Of course, this is not completely possible in practice.
Regarding masochism, if you do it retrospectively, it is. But performance would greatly improve if every modeller would make such an analysis first. However, it is not possible for most objects to predict how far from the track they are placed. I am curious whether TSW will address this somehow, maybe by only loading distant LODs when it knows that the nearest one will never show? More likely, DTG have their in-house rules and don't care what others do.