Let's say Person A does this with $100, once. Person B is in cohoots, and they are going to split the money (assuming both kicked in $50)
Best case scenerio: B wins the roulette coinflip. Total gain, $100.
Other cases:
1) B loses the coinflip. A complains and is reimbursed. +$0
2) B loses the coinflip. A complains and is denied. -$100
3) B loses the coinflip. A complains. The casino reviews the tapes, and finds a connection between A and B. Both are banned from the casino. -$100
4) Same as 3, except that A and/or B are arrested. -$huge. Bail, legal fees, lost wages, fines, jail time, beach-front beatings, etc.
The way I see it, there's only 1 scenerio where they are going to win any money, and there's only a less than 50% chance of it happening (Roulette's house edge). If it was equally likely that the house would reimburse as not, then you're looking at:
EV = .5 * 100 + .25 * 0 + .25 * (-100) = 50 + 0 - 25 = +25.
Now let's figure out what the minimum chance of being wiped out (case 4) is to make this profitable, assuming a 50/50 shot of getting their money back is, assuming the cost of scenerio 4 is $100,000 (legal fees and a few years of lost salary)
EV = .5 * 100 + .25 * 0 + [(X * -100) + ((1 - X) * (-100000)]
EV = 50 -100X -100000 + 100000X
EV = -99950 + 99900X
So, 99900 / -99950 basically means that there would need to be at MOST no more than a 0.0006% chance of being prosecuted. This percentage could be reduced if the players increase the "stolen money". If it was done with $1000, it would be:
EV = 500 - 1000X -100000 + 100000X
= 99500 + 99000X
= 0.006%
So as the bet size increases, the minimum chance of "Case 4" being OK increases. Except that I'd say that the higher the bet, the less chance there is of a casino refunding the money. It's plausable that someone swipes a green chip. I'm sure the casino wouldn't make a huge fuss over a green chip, given a one time occurance, good tape, and a convincing acting job. In fact, one can more probably see a chip-swipe HAPPENING at a low-stakes table.
But to be in the position where $1000 or $10000 is even in play-- and not have it suspicious to begin with-- not as likely.
All this is assuming a one-time deal. If this grift were run multiple times, with the "world" having memory of each previous attempt, I'd hazard a guess that the chance of Case 4 will expotentially rise to 1. It's fishy to begin with, and having it happen multiple times makes it fishier. Even if it's at different joints, it's only a matter of time before some dealer/critter/security person/owner, etc is talking to another of the same from a different casino, and they both have the same story to tell. "This guy did this..." "Yeah, that happened to us to...". From there it gets passed up the ladder, spread, boom.
Now if this happens when the two players aren't in cohoots, and the B is actually malicious-- well, then that's a civil matter. The casino MIGHT refund, might not, won't charge. But I'm sure A will be having some words with B. =)