This reminds me a lot of typical Newbie queue games (which often come down to a 1v1 in day 3 lylo; this setup just forces that directly).
That said, I don't see the forced 1v1 actually helping town much, especially given that both participants are chosen by scum. If there was a clear 1v1 in the setup already (not that that's likely to happen with no power roles…) the scum will just choose it and duplicate the information town had already. If not, town's chance of surviving a day 3 lylo increases from 40% to 50%… and that's the only change in the setup, as scum will just kill the confirmed townie overnight.
FWIW, there's a 5/9 chance of reaching a day 3 lylo with random lynching in a 7:2, and a 1/3 chance of winning if you survive it, so the EV improvement over a 7:2 vanilla game should be 1/3 of 5/9 of 10%, or 1/54 (about 2%). That's improving the theoretical EV from 29.8% to about 32%, which is still far too low. (For what it's worth, we've had an effectively vanilla setup in the Newbie Queue before now – F11 subsetup D – and the observed win rate in practice was 31.94%, very close to the theoretically predicted 29.8%. So I wouldn't expect the theoretical win rate and practical win rate to be particularly different for this setup.)
Making it so the duel survivor (if town) can't be nightkilled that night would increase EV by about 6% instead. A little more respectable.
Going to 2:9 (and keeping the two mislynch rule as is, rather than going all the way to lylo) bumps the EV to 46.12% (vs. 35.21% for Vanilla), pretty reasonable. It guarantees town will get down to one scum left at some point (day 1 successful is 1:8, day 1 mislynch and day 2 successful is 1:6, successful duel is 1:4, failed duel is 1:2).