Not when you consider the underlying assumption, which is that the horses are all bet in the exotics in exact accordance with their win odds.

The more horses involved, the further the variation can be, because there's a multiplicative effect on the estimated payout for each horse whose odds are different in the exotics than they are in the win pool.

The estimated payout for [i]vertical[/i] exotics is further complicated by the fact that horses don't run 2nd, 3rd and 4th at the same rate as their win odds would indicate (the distribution range is tighter), which is why the computer guys use a discounted Harville method rather than straight conditional probability.

The proof that the assumptions are valid though is in the long-term results, which on average are very close to the actual payouts. I'm actually surprised that the long-term results are as close as they are, given that there's no direct link tying a horse's odds in the exotics with that of the win pool, it just works out that way on average through the "wisdom of the crowd".