It seems that after all the trouble of a difficult game, at the very end, sometimes blunders happen that are inexplainable normally. The side to move is marked by asterisk.
Let's start with Kortschnoi unwilling to win a won game.
Azmaiparashvili - Kortschnoi *
Moscow (rapid) 1995
Black can keep his pawn with 1...Se5 and 2...Sg6, or 1...Sh6 and 2...Sf5. Kortschnoi, who played for the World Championship some years before, finds a third way to keep the pawn: 1...h3?? 2.Kg3 h2 3.Kg2 Kxa4 4.Kh1 and a draw was agreed. No explanation can be offered for this, except that he forgot the most basic endgame rules for a second.
Chandler - Susan Polgar *
Black is already in dire straits, and her move 53...Sh6(!) wouldn't have changed that if Chandler wouldn't have played checkers in his mind for a moment. As it was, however, after 54.gxh6+?? Kh8! it was a draw which was agreed two moves later. In checkers you have to take, but in chess...
The following blunder at least can be explained since pawn endgames are one of the most difficult types of endgames, and of course the only ones where even with four pieces a huge load of studies does exist. So it is no wonder that even grandmasters fail here.
Malakhov * - Najer
We don't know who showed the position first - Moravec's 1940 study (below) is already marked with the indicator for being anticipated in Harold's database - but if Malakhov knew it, he could have easily won the game here. Maybe he knew it and his issue was that the position was not exactly like the one in the study? Anyway, he tried the direct attempt by 1.Ke5? Kc5 2.f3 Kc6 3.f4 Kd7 4.Kf6 Ke8 5.Kg7 f5! 6.Kf6 and a draw was agreed. The win was so easy... to miss!
Ceské Slovo 1940
White to move and win
1.Kb1!! Kg2 2.Kc2 Kf3 3.Kd3 Kf4 4.Kd4 Kf5 5.Kd5 Kf6 6.Kd6 Kf7 7.b4! (but also 7.Kc7 or even 7.Kd7 would win here) 7...Ke8 8.Kc7 b5 9.Kc6 Kd8 10.Kxb5 Kc7 11.Ka6 Kb8 12.Kb6 wins. Knowing this it is easy to see now that Malakhov indeed could have won by 1.Kd4!!, reaching a position from the study (shifted and mirrored). Play could continue 1...Kb5 2.Kd5 Kb6 3.Kd6 Kb7 4.f4 Kc6 5.Ke7 f5 6.Ke6 Kd8 7.Kxf5 Ke7 8.Kg6 Kf8 9.Kf6 and white wins like in Moravec's study.
Batuev - Simagin *
"Everything" wins, so it is hard to imagine that black can even lose this. However, he can: 1...e2?? 2.Qg1+ Kd2 3.Qc1+ Kd3 4.Qc3 mate. A moment of being unconcentrated can throw away a game.
Chigorin - Schlechter *
Of course, black is lost, but he manages to set up a final trap:
1...Qc7+ 2.Qb6+? Ka8!!. Draw. Either white stalemates or repeats moves by 3.Ka6 Qc8+ 4.Ka5 Qc7!
Examples taken from:
John Nunn: Understanding Chess Endgames, 2009
Mark Dvoreckij: Die Endspieluniversität, 3rd German edition 2006 (is equal to the 2nd English edition)
Wikipedia (several authors): Blunder (chess), 2010
Harold van der Heijden: hhdbiii, 2004
As a conclusion, I'd like to give an explanation why the following famous world championship blunder is not included here as an example.
Bronstein * - Botvinnik
World Championship 1951, 6th game
Bronstein played 56.Sd8 e3. Now we can often read that he blundered by 57.Kc2?? Kg3, overlooking black's reply. However, I once read a simple explanation for that. Botvinnik indeed saw the draw by 57.Se6+, but was in thought and touched his king undeliberately. Since he had touched it, however, he thought to be required to move it and played 57.Kc2 now. No blunder, but a fingerfehler. Sadly I forgot where I read that...
Readers are invited to post more examples!