"Cheese" is a pejorative, loose term used in the starcraft community in a variety of ways.
In the strictest and most accepted sense cheese refers to an early game strategy which is designed to surprise the opponent and which can be easily countered if the opponent knows that it's coming. Almost always the economy is sacrificed in the process, sometimes it's given up entirely by bringing all the workers alongside the army to attack.
Some of the paradigmatic cheese strategies in this restricted sense are:
- Cannon Rush : as Protoss skipping a gateway and building pylons and cannons in the opponent's base, often outside the opponent's vision, in the early game
- Proxy Gate/Rax: building the first production facilities close (or inside) the enemy's base.
- 6 Pool: creating the Spawning Pool as the first building, skipping initial drones, and then attacking with 6 zerglings.
These kinds of cheese strategies are often strongly disliked because they can even the playing ground by making strategy and game knowledge count very little. The mechanical skill is also less important. In this way a weak player can specialize by learning to play a single strategy with low mechanical requirements to beat a player he would never be able to in a normal game. Cheese is also often said to be easier to execute than to defend. Because of these reason this kind of cheese is often felt as "unfair".
Normally Cheese strategies in this sense end with one of the two players winning or having a very large advantage, even though rarely the game can end up being suprisingly even.
Sometimes the term Cheese is used to indicate an aggressive strategy which develops later (say from the late early game to the early mid-game) which has about the same high-risk high-reward ratio as the earlier strategies and which tries likewise to surprise the opponent. The border between this kind of cheese and an All-In strategy is pretty narrow, often depending on the player's perspective. If one had to try to say what the difference is, one might say that an All-In strategy is not "easily" countered even if the opponent knows that it's coming, even though the suprise would be a great benefit. An All-In takes advantage of the lack of scouting, but doesn't require to surprise the opponent to the extent a cheese does.
One example could be a proxy Robotics Facility strategy.
Note that this second kind of cheese doesn't level the playing field nearly as much as the very early game strategies do.
In a third, even broader sense Cheese is used to denote any strategy which has a high-risk high-reward ratio and which is weakened if the opponents scouts or reads it correctly. Thus one might talk about economic cheese for strategies which forgo defense early on or expand beyond the capacity to defend, expecially if the expansion is hidden (located away from the opponent in an unusual place). Even 20 minutes in the game one might talk about an hidden undefended expansion being "cheesy".
In this last sense the use of the term cheese can easily become slippery since many of the normal aspects of the strictest definition are missing, particularly for the "economic cheese": the reward can often (e.g. very fast third base as protoss ) be quite subtle, it's often pretty delayed, therefore there is no prospect of winning outright, so the opponent has a lot of time to scout, economy is not sacrificed (of course), often the opponent cannot punish "easily" in a fast way even if he scouts and, most importantly, the cheesing player has to be about the same skill as the opponent to win using it (since the strategy gives an advantages only in the later stages of the game), and this kind of cheese is not really easier to execute than it is to respond.