Chess.com has an interesting article that looks at performances in some of the greatest tournaments in history. 12 events were chosen, starting with London 1851, and players estimated elo rating were calculated using the chess.com CAPS system.
Of course the earlier events had reasonably low performances (due to a lack of accurate play), but once into the 20th century there were some players getting quite high estimates (eg Capablanca 2726 at St Petersburg 1914). One interesting feature was that tournament winner often did not have the highest estimated rating, with some like Averbakh having the top TPR of 2759 at the 1953 Candidates, despite finishing 10th, and Euwe having the 2nd highest, despite coming 2nd last!
It is worth noting that the margins between the players became a lot smaller in the 21st century, which isn't a surprise as modern players train against the very tool that is now judging their play. By 2013, the CAPS scores had risen by so much, that 7 of the 8 players at the London Candidates played above the 2700 level.
The article itself is here, if you wish to have a look at it, with the disclaimer that models like these do reward accuracy over creativity (which is noted in the article itself)
