What makes Swindler high-variance, it seems, would be
1) Based on shuffle luck, it can have a huge effect. If I swindle your first five-cost card into a duchy (or swindle away your potion in a game with critical potion-cost cards, etc), whereas your swinder hits my estate (discarding it from the top of my deck and replacing it with an estate, helping me instead of hurting me), that can make or break a game.
2) It comes out immediately. You can't try to "be better" than the other guy by getting to Swindler first, and you can't really try to prepare your deck to be hit by a swindler, because it's an opening card. Comes out too early for you to really do anything about it besides hope that your swindler hits better cards than theirs.
The way variance seems to be typically used here is on a whole-game level. In a game, usually the better player is going to win. However, cards that make the game more random will tend to make the worse player win more often - hence, increasing the variance in the result. That's not necessarily the same as cards that have a very different expected value from turn to turn. (KC, for example, can facilitate *very* different turns - you might use a KC->KC start to have a 100-point mega-turn, or you might draw it dead. But typically, since setting up the KC->KC mega-turn takes a lot of setup, it's the better player that will get their KC-KC->support set up first, making the variance in the game result due to the presence of KC pretty low. )
I'm definitely looking forward to that article WW mentioned, or to playing some Possession games with people... it feels to me like Possession is high-variance in all those respects; you might get a province turn or you might discard a 2-coin hand for them (and vice versa), but apparently better players than I disagree, so there's something I'm missing.