"Water hardly ever freezes at 0 °C," says Brownridge. "It usually supercools, and only begins freezing at a lower temperature." The freezing point depends on impurities in the water which seed the formation of ice crystals. Typically, water may contain several types of impurity, from dust particles to dissolved salts and bacteria, each of which triggers freezing at a characteristic temperature. The impurity with the highest nucleation temperature determines the temperature at which the water freezes.
This work is unlikely to end the Mpemba debate. Jonathan Katz of Washington University in St Louis, Missouri, says he is sceptical. According to Katz's own theory, heating raises the freezing point of water by driving off solutes such as carbon dioxide. This means that heating the water actually increases the chances that it will freeze first, unlike the more random outcomes suggested by Brownridge. "Perhaps he has found an effect of supercooling that resembles Mpemba," says Katz.
And here's how the effect was named rather than casually observed:
In the 1960s, the effect came to the notice of modern science when a Tanzanian schoolboy called Erasto Mpemba told his science teacher he could make ice cream faster than normal by putting a heated mixture in a freezer. Mpemba was the laughing stock of his class until a school inspector in Dar es Salaam repeated the experiment and vindicated him.
This is nostalgic for me-- one of the first things I read on usenet was a long discussion of why hot water freezes faster than cold.