The speculation has been that Google thinks it will make back what it spends on an SF Wi-Fi network in more advertising sales: The first supposition is that by adding a Wi-Fi network citywide that Google will increase overall Internet usage. Because it already owns the minds of a large percentage of Internet users in general, branding their putative San Francisco-wide portal won't shift lots of users from other search engines or outlets. They have to rely on adding clicks if they're trying to justify it in part through additional income from advertising clickthroughs.
Google doesn't have a per-se average clickthrough price. You can pay a nickel a click for unique and unpopular keywords, but those aren't delivered very often for that reason. Keywords relating to asbestos litigation can cost $60 or more a click.
I'm going to estimate that for Google to build the low-broadband-speed ubiquitous network they propose for San Francisco that they'll have spend $5 million to $8 million. Philadelphia's network will be more expense because they have twice the population (over 1.5 million versus SF's 750K) and nearly three times the area (135 square miles versus 46 square miles)--and because Philly wants a higher level of minimum bandwidth. They'd like 1 Mbps in each direction; Google's plan promises 300 Kbps.
If Google can add a few million incremental clicks a year, it's possible that the network will be entirely paid out of ads. But that's not what they're aiming for--they don't really need incremental users to drive delivering more ads and clicks. They're aiming to move more advertising dollars out of the devastated newspaper business in the city and suck more life from telephone book display advertising. National advertising in the U.S. comprised $45 billion the first half of 2005; local advertising, $26 billion.
Because Google will run the network, they can deliver ads targeted to the city block for folks using their Wi-Fi network without knowing anything about the individual consumer, as it will be entirely based on the Wi-Fi network not consumer characteristics. I imagine Google views this as a massive experiment and money well spent.