I think we're in substantial agreement as to Paypal desiring to shift losses due to fraud or non-performance onto the merchants. Since they desire to do this, they will not accede to Mailpile's desired world, where Mailpile gets 100% of the money on day 1 and then Paypal can seek recovery on day 120 or 365 if a chargeback comes in. Many HNers wish we lived in Mailpile's desired world 100% of the time regardless of merchant risk profile, but we don't, which was the general thrust of my comment.
In terms of predicting Paypal's disposition on this particular case, I think it's a little trickier than a 90 day rolling reserve, because the product doesn't exist and shipping is binary rather than mapping to the rolling window of binary events like a physical goods business. If they were selling trendy iPhone cases and had bitten off more orders than they could presently chew, then an X% rolling reserve for Y days would likely be sufficient to insulate against the risk of some orders not shipping. Software businesses can fail on day 91 in an explosive a fashion as they can on day 1, and it appears from the report of Paypal's communication that Paypal is aware of this and is disinclined to do a fractional rolling reserve absent shipping. One would assume that if they ship this business goes to fairly low-risk instantaneously [+] and a reserve is probably unnecessary at that point.
Ultimately that's up to Paypal rather than either of our guesses.
[+ Well, it would be low-risk with respect to new orders instantaneously after shipping. It might not be low-risk with respect to pre-orders for a while yet, as customers could suffer from buyer's remorse, be dissatisfied with the product as shipped, not remember the transaction, or engage in friendly fraud, at rates substantially higher than transactions with lower delay between purchase and delivery.]