I just read the following article on chess.com about a game played by Lev Polugaevsky in 1979. Here it is.
To make a long story short, Polugaevsky plays an aggressive pawn push on move 17 and secures a winning advantage shortly after because of the opponent's mistake. The article shows the main variation(s) that should have been played (second diagram w/ move list in the link), which runs through move 36, and Polugaevsky claims to have calculated the majority of it over the board during his 1-hour think on move 17. A large part of that 19 move variation is forced, but some of it isn't, and there are also several hanging pieces, checks, sub-variations, etc, that would make calculation more difficult.
It's simply amazing to me if he really did calculate almost all of that OTB with the clock ticking.