The
Councilroom list of popular buys shows Potion as one of the worst cards in win rate; it's bought half the time and has a win rate of 0.94 with the card and 1.06 without it.
I believe this is an artifact of the way the game is played on isotropic, which takes a random selection of kingdom cards; there would be much less of a split if you played with a minimum of three Alchemy cards (and thus usually have 2-4 Potion-cost cards rather than just one). This also explains why Potion is bought only half the time.
Potion is good in some sets, bad in others, depending on which Potion-cost cards are available. If there are several Potion-cost cards, then Potion is likely to be a good card, and everyone will buy it. If there is just one Potion-cost card, the weaker player is more likely to overvalue the card and buy a Potion. Note that even Possession is break-even on win rate; if Apothecary and Alchemist are there to make Potions useful, then the player who uses them on the way to Possession will do very well, but if Possession is the only good Potion-cost card, then buying a Potion will slow down your deck and you will be too far behind when you get your Possession.
The only very good Potion-cost card is Familiar, and even it scores much worse than Witch and Mountebank. This is probably because of the delay in getting Familiar into the game; if only one player has Witch, he may have bought it on a 5/2 opening, and that made it hard for his opponent to get to 5 until a lot of Curses were already out. In contrast, Familiar cannot be gained until the second time through the deck.
The other two Potion-cost cards which are better than average are Vineyard and Apprentice. Apprentice is interesting; it's better than the similar Salvager, probably because Alchemy decks can make good use of the actions from Apprentice in a big hand. Vineyard has cause and effect reversed; if you are buying Vineyards, that is because you have a deck which benefits from the VP from Vineyard, and usually that means a strong deck.