You are also assuming the benefit behaves in a stepwise fashion. I.e. you either get the toxicity or you don't. Perhaps there is more nuance to it that makes analyzing it from a simple NNT calculation not explain the full story (non-stochastic situation that shows increased benefit with decreasing rectal dose):
"At 3 years, more men in the control group than in the spacer group had experienced a MID decline in bowel QOL (41% vs 14%; P=.002) and urinary QOL (30% vs 17%; P=.04). Furthermore, the control group were also more likely to have experienced large declines (twice the MID) in bowel QOL (21% vs 5%; P=.02) and urinary QOL (23% vs 8%; P=.02)."
So maybe think about how we are determining toxicity grading and QOL and why is there such a dramatic improvement in the QOL stat but a more modest but still solidly significant improvement in graded toxicity ((9.2% vs 2.0%; P=.028) and grade ≥2 (5.7% vs 0%; P=.012))
In other words, I don't fully buy the "if you put a spaceOAR in 20 people then exactly 1 person will benefit and the other 19 will derive exactly zero benefit at all." I also don't think we have enough data at this point to confidently say that the benefit is "at best" 5%. In fact, in the 2017 update showed an absolute benefit of 7.2%. And in the 2015 paper you linked, of 222 men, they reported zero " device-related adverse events, rectal perforations, serious bleeding, or infections." So you can do the math on that. If you believe your NNT analysis of 20:1, then given 0 of 222 harmed, it would seem that NNR is much higher than 20:1, no?