In practical terms, little to no money is likely made from that float - it's more than likely that any money they could make, would be offset by the losses from chargebacks/fraud/mistakes and the operational risk costs of the transactions (for example the risk that while the float is being held one of the two banks declares bankruptcy). I would suspect as well that the float money would be limited in terms of what the bank can use it for in terms of short term investments as it might well fall into regulations regarding the bank's capital reserve requirements as the money doesn't technically belong to the bank at that point.

I'll say though that retail banking is not my area - I work with derivatives and risk.

For what it's worth, I've got no beef at all with someone trying to do ACH better. I just think that for Dwolla to claim that they'll be able to do so is either massive hubris or massive naivety, neither of which I want in my payment processor. They're a tiny company backed by a small company, and that does not bode well for their ability to deal with operational risk when they're talking about these sort of ventures.