Perhaps I'm in the minority here given the comments thus far, but I've always been of the opinion that equivalent work should be compensated equivalently and commensurate with the value of output generated for the company.
There is some value in the firm having a culture of unity, etc. but bigger companies are going to just negotiate with individuals as buyers in a labour market similar to you buying a car warranty contract.
Whether you believe WFH is equal in value to on-site is another matter. Google apparently does not hold this view.
Regardless, in those cases I’d still expect a competitive market to reach some price convergence, holding all other factors constant — especially when the seller knows that a buyers willingness to pay == $x and != $x-d for the same utility (say, from already getting paid $x repeatedly for the same work). If I am mistaken I’d definitely welcome the chance to adjust those expectations.
This is the crux of the problem, of course. There is no convergence because the only factor isn't that the SF engineer is competing against engineers worldwide. The price you have to pay is the minimum to prevent someone from doing the other thing. And so, even if Google is the only one in the Bay to offer higher to work there, you need to pay as much (in some terms, not just monetarily) as Google to get an SF engineer.
That makes sense to me. Local large employers can have an effect on local hiring costs, even for employers not stationed there.
It's not Baumol's Cost Disease, but there is an analogous mechanism there.