Rio Grande says that this is a 2–4 player game (as does FoxMind). However, it does not play well with only two players, nor is there any barrier at 4 players.
The problem with 2 player Corsari is the the odds favor sailing on the first turn over half the time! (Well, over a third of the time if you count points instead of skunks. Still very often.) Maybe if you often find yourself wanting to play with only 2 you could cook up a way to mitigate this, such as disallowing sailing in the first round, disallowing adjacent cards in the tavern with the same color, etc.. Or perhaps it doesn't bother you, in which case never mind. :-)
It seems to work just fine with 5 players. 6 and 7 player games are technically possible, but this makes the supply very small and rounds very long. I have played with a bunch of people who think that it is a fine 6 player game, but I think that with 6 you have to sit around for too long between turns. This is especially annoying when you don't even get a turn in a round. I haven't heard anyone advocate for 7 players... yet. You can deal enough cards for 8 players, but this leaves 1 card in the supply (or a few more if you cut back on the tavern) and so is clearly no good.
I have blindly generalized the pattern for larger numbers of players. It's not clear that larger games need 10+ card taverns, but it seems to work fine. Cap it at 9 cards (the number Rio Grande and FoxMind give for 4 players) if you think it makes a favorable difference.
Rio Grande says that players are eliminated as they reach the maximum score. I don't like this for two reasons. First, it is socially awkward, since the eliminated players have nothing to do. Second, the game does not play well with 2 players, which is where this method leads. My method solves these problems and also leads to some interesting cross-round strategy.
Rio Grande suggests playing to 101. That's fine, but can make larger games end very fast. 5-8 rounds seems to be a satisfying length for most players. My suggestion of 30×#players, minimum 100, seems to work decently. 40×#players seems like too much. In my testing, this often leads to one player getting an unassailable lead with several rounds to go, which is annoying unless you enjoy trying to sail with zero.
In contrast, FoxMind's I Go has completely changed the scoring by simply having players win rounds (the sailor if not skunked and all skunkers otherwise) and counting the number won. This significantly alters the strategy and arguably reduces the game's strategic complexity. For instance, with our scoring, you may choose to sail late in a round even if you think there is a high chance of being skunked — because you know that even if you are skunked you can't get more than 10 plus a few times your small point total — in an attempt to catch another player with a large number of points.
Rio Grande (and FoxMind) would like you to seed the discard stack with a card. The problem with this is that the first player has a huge advantage, especially in smaller games, because of the possibility of sailing before anyone else has had a chance to do anything. (On average, each player can drop about 10 points on the first round.) Seeding the discard increases this advantage even more while not doing anything positive for the game, other than, I suppose, indicating where the discard pile is.
In concrete terms, my simulations have found that the discard seed allows the first player to drop an average of 0.95 more points that he could otherwise. To look at it another way, the probability of being able to announce a score of 17 or less on the first turn (which gives better than even odds of not being skunked even in a 5 player games) is 16.9% with a discard seed and only 14.9% without.
Rio Grande does not specify what happens if a player sails with a non-zero score and two other players achieve zero as a result. They also aren't entirely clear whether a non-ending player can win by sailing with zero. I've cleared all this up.
Rio Grande does not clarify whether one is obligated to choose two colors when one sails. Sometimes it could be an advantage to sail with only one (or perhaps even zero in very extreme cases) in order to avoid letting other players dump points. This has been a contentious issue, but we have settled on allowing people to name fewer than two colors since that is strategically more interesting.
Should skunking players be allowed to choose crew colors that don't give the lowest possible score or otherwise raise their scores so as to give the ending player more points? This isn't clear in Rio Grande's rules (and is irrelevant in FoxMind's rules). I've taken the clarification least likely to cause arguments as well as that which reduces computation for the skunking player. This is somewhat contentious.
FoxMind tells you that the sailer becomes the dealer for the next round, regardless of whether they "won" the round. This is a substantial disadvantage, and I guess was added under the assumption that the sailer usually wins the round. We have not adopted this rule, but favor passing the dealership around the table as per usual.