However I think he made the same mistake I struggled with too, which was thinking that if the bankroll is -EBG that would allow a player to be +EBG. Fortunately (for investors) this isn't the case, so there's no real abuse avenue.
Can you check where I'm going wrong with this? Here's how I looked at it:
Let BG and PG be two random variables, BG is the house's bankroll growth and PG is the player's bankroll growth (in absolute numbers). Then:
BG + PG = 0 (money only moves back and forth between the house and the player, so a loss for one is a gain for the other and vice versa)
E[BG + PG] = E[0] (taking the expected value of both sides)
E[BG + PG] = 0 (expected value of a constant is that constant)
E[BG] + E[PG] = 0 (by linearity of expected value)
E[BG] = -E[PG]
Which leads to the player's EBG being the opposite of the house's EBG, so if the house is at -EBG then the player is at +EBG and vice versa. Perhaps by EBG is meant the median bankroll growth and not the mean bankroll growth? In which case the linearity condition above wouldn't apply.