I seem to see commentators mention "cheese" with regards to Starcraft 2, but I'm not quite sure what they mean.
If a particular build/tactic is called a "cheese", does that mean it is a considered to be uncounterable? Or does it just mean that it is considered to be a tactic that is "too easy" with regard to its effectiveness? Or is there some other more subtle meaning that I am missing?
The term "Cheese" comes from the adjective "Cheesy". The not-so-subtle innuendo there is that a victory via so-called "cheese" did not come from the players own skill, but was merely because the loser was unprepared for the "cheesy" tactics that lead to their loss.
What qualifies as "cheese" will undoubtedly vary from player to player, but generally the term refers to exploiting system mechanics in a way unanticipated by your opponent, or otherwise not balanced around.
Many people dislike cheese because it is hard to predict and counter if you are not prepared for it.
Several Examples:
I would have to strongly argue against those who say that "cheese" is an exploit in the game.
A cheese strategy is usually an all-in strategy.
Reaper rushes, zergling rushes, roach spam, proxy gate rushes are something lots of people consider cheese strategies.
If a cheese strategy is countered, then the game is usually easily won. Cheese builds in SC2 have very distinct openings and you should learn to scout them and be prepared.
In short, cheese does not work against competent opponents. The only reason they work is because the opponent was not expecting it. However, being a competent SC2 player means that you should always know what your opponent is doing.
While the first answer has some good points, most cheesy tactics are not "exploiting system mechanics." To call it that would imply it needs to be fixed.
The simplest definition of cheese is any tactic that is considered to require more luck than skill. The example of early proxy rush is a good one. The reason is, you can do proper scouting of the enemy base and the area around your base, but if the enemy got lucky and you just happen to miss his proxy either because your search path just missed his building, or missed him timing-wise (you search a place just moments before his worker gets there). Same goes for something like Void Ray rush. It's very early in the game and the player will try to build it off in a place where they don't expect you to scout. If you don't see that starport early enough, that first void ray may be enough to end the game for you.
It's more about taking advantage of a flaw in the system to get a cheap win. Kicking someone's butt with a reaper-rush, which takes advantage of the fact that the defensive buildings are weak, is cheap. You're not showing skill, you're just exploiting a weakness in the game.
Edit: Reaper rushing is a dated example now, since they've done a good job of making that unit balanced.
Cheese involves an attacker employing high-risk, high-reward tactics in order to win. Such tactics are typically difficult to detect by the victim and often leave the attacker extremely vulnerable. Thus, if the victim is able to preemptively detect that their opponent will be using cheese, the victim is typically able to formulate a strong counterattack.
Cheese receives its negative connotation from those who believe that the cost of preemptively detecting cheese is too high. However, in the case of Starcraft 2, this cost is typically very minimal, usually requiring the use of 1-2 workers and chaining scouting commands.
My feeling about Cheese is that is just as legitimate as camping in other types of games, and irritates other players for much of the same reason. I like Daryl's explanation that it incorporates luck more strongly than the more stable builds many players prefer, but I'd like to extend that it works principally by exploiting to maximum advantage the asymmetries between the opposing forces.
A particular feature of cheese is that it works well before the opponent is able to get a reasonable sized, balanced and stable army, because it exploits the compromises players have to make in the early part of their builds. By mid-game, those weaknesses will be covered by the composition of the opponents army.
For this reason, you see cheese play mostly in non mirror matchups. There are more chances for asymmetric advantages in a TvP, PvZ or ZvT matchup than in TvT, PvP, and ZvZ. That said, there are still some interesting things that can happen in mirror matchups that would typically count as cheese.
As a particular example, I'll compare TvT where the defending player is using a 1/1/1 build. A cheesy terran player might bank on the opponent making marines out of their barracks until the factory is up in order to defend against early aggression. Unfortunately for the defender, marines aren't very good against reapers. If the cheesing player arranges for a handful of reapers to arrive before any addons are down on a factory or barracks, it will be very hard for the defending player to hold it off.
On the other hand, the cheesing player could be very unlucky and the defending player got a tech lab on his barracks immediately. A very small number of marauders can hold off many reapers because reapers do so little damage to armored units.
How about this:
A simplistic tactic designed to get an quick win that unreasonably demands the opponent to adopt very specific and unbalanced countermeasures to survive. Consequently, it is not usually very fun to play against (unless one is expecting it, perhaps)
A little bit of trivia:
The Korean term for a cheesy strategy is 'strategic play' (as versus standard play).
I would say cheese would be proxy anything, or cannon rush. 2 gate proxy in their base, 2 rax proxy, cannon rush, etc. These are all strats that dont take much micro nor macro, and are hard to defend against early game and don't really get scouted either.
d5. Black comes out of the opening way ahead. If you are still unsure, ask a new question on the boardgame SE; I'd be happy to go into more detail there.