The idea that savings from physical distribution factored in at all is laughable.
....
Snow Leopard was already being sold at a great loss - There's no way $30 a pop comes close to recovering the development cost. They just don't care, because OSX is developed solely to drive hardware sales - not be a viable business on its own.
Snow Leopard and Lion's potential profits are completely different stories. I'll argue that Lion's development will not be a loss on it's own revenue basis, and back it up with numbers:
It's hard to guess how many upgrades of Snow Leopard sold and how many Lion will sell, but this chart should help:
http://www.ifoapplestore.com/stores/charts_graphs.html
It looks to me like somewhere in the ballpark of 5 to 6 million CPUs were sold between the Intel Mac debut (late 2005) and the release of Snow Leopard (late 2009). So let's guestimate it at somewhere between 140 and 180 million in revenue for the Snow Leopard upgrade, and not even include the revenue of new users drawn in to the hardware/software combination by features like Exchange integration.
I wouldn't say it's "laughable", but it
is chicken scratch compared to the nearly 40 billion Apple did in 2009.
Lion is a little different - the number of Upgradable Intel CPUs sold to date is much bigger. Almost all CPUs sold since 2007 and many earlier ones are potential sales. CPU sales from 2007 are somewhere around 9 million, resulting in a minimum 270 million in potential upgrade revenue.
With 2 years of development since the Snow Leopard release, if you payed all the engineers 1 million dollars per year each, you could hire 135 engineers. Obviously, that's a wee bit high in salary and a few too many cooks, IMO.

Any way you slice it, there's profit.
What's really interesting is when you compare the ratio of profit to employee between Microsoft and Apple: Last I checked, Apple's ratio was
double that of Microsofts.
Dave