So here is another one: Steinitz v von Bardeleben, Hastings 1895
Steinitz: 0 inaccuracies, 0 mistakes, 0 blunders, 7 average centipawn loss
von Bardeleben: 2 inaccuracies, 3 mistakes, 0 blunders, 31 average centipawn loss
"As Steinitz demonstrated immediately afterward, there is a mate in ten moves which can only be averted by ruinous loss of material; analysis follows: ...Kh8 25. Rxh7+ Kg8 26. Rg7+ Kh8 27. Qh4+ Kxg7 28. Qh7+ Kf8 29. Qh8+ Ke7 30. Qg7+ Ke8 31. Qg8+ Ke7 32. Qf7+ Kd8 33. Qf8+ Qe8 34. Nf7+ Kd7 35. Qd6#"
I think any of the superclass Grandmasters today would be happy to play as game as brilliant as this one by the 59-year old ex-World Champion.

1932=middle of 20th century?
Anyways I don't think that centipawn average loss is such a great indicator of how good a chessplayer is; good is a term that's very hard to make objective.
Edit: looking through the Dake game gives a more legit reason that top players today are better than they were: Be6 deprives Black of the e5 break against the c5 push and so is bad on Black's part. I couldn't imagine a world-class player playing this, but Anand played it in a rapid in 2009 against Nepo, and Navara won with it in 2014. iirc theory still frowns on it now.
First third of the 20th century!
"Average centipawn loss" is essentially a measurement of the human players moves against what the engines scores as best. Engine scoring is unaffected by nationality, age, favorites, fads, trends, or fanboyism.