Well, I tried setting Z backlash to 0.02, which is a full step, and running the program in the air, and it didn't lose position.
Before that I tried 0.01, because that is the size of a step while microstepping, but the position losing still occured, although it was quite smaller.
So it seems like the backlash setting solved it, but I didn't have time for running the program long, so I might be wrong.
Do you say that backlash has nothing to do with it and these are false results? If so, then I'll have to buy another motion control board to eliminate that failure possibility.