Are you aware by any chance of how these checks are conducted? Are the checks made of real-play or just a fun-mode basis? This could be important as the allegations are that firstly (and this HAS been proven in both the Spielo affair and on some iffy software mentioned in the rogue section) the game reacts differently to real-money play by being 'stake-sensitive' and secondly the real-play game differs from fun-play. Both are forbidden.
So, it's all very well testing a game's programming by pumping it for millions of results to establish there is a fair, reasonable and expected spread of outcomes, i.e. on BJ and roulette - that just means the basic RNG and overall TRTP function is being confirmed. But do these tests actually use random bet amounts/stakes and analyse if the RNG then produces a different result? This is where some people allege the disparity is - winning is being 'managed' somehow.
I don't know exactly how these tests are conducted, but I do know that the testing houses aren't fools, and that they examine the entire chain of the game round.
The most likely scenario is that the game result server, the server that spits out the gaming result, is not aware of the difference between a "for fun"-mode and a "real money" mode. It makes very little sense to have the same piece of software handle both the monetary transaction AND the RNG/reel-set win management code.
The way I think it is being set up is that one server handles the money (checking weather the the cost of the bet can be covered, stc), and then a request is sent over to the RNG-server that handles the actual game result.
The reason I believe this is the case is, not only from a code design point of view, but also from a "real world data" point of view: If you remember some issues that we have seen reported on here, players say "It runs fine in Free to Play mode, but it takes a looong time to get the spin results when playing for real". I've had that happen quite frequently, and I am pretty sure that there's been a number of threads talking about this delay.
In the free to play-scenario, the money-server doesn't have to do anything, so there is no delay, but in the "money"-scenario, the money server has to store play sessions into a database, keep track of all the play rounds, etc. (In "For Fun"-mode, the game history isn't available for the games, that only exists in "money"-mode, which means that a server does a lot more in the "Money"-mode. )
I may be wrong, naturally, but this is what makes the most sense to me, from observed delays during game play, and as a software engineer in general.