I'm guessing that the higher salaries are for people who are experts in (did their dissertation on) extremely narrow subfields of subfields which Google has an intense (but probably only capricious) interest in-- they want THAT person because they're one of a dozen on earth who specialized on that particular thing. This would drive the salaries up quite a bit, but only as blips. From the outside, this would also explain the "seemingly random" criteria used to select people-- it's not random, just very focused on an otherwise obscure/esoteric item.
Edit: Worth clarifying is that the cases I know of first-hand for this got their PhDs in "bioengineering" or something similar, not a traditional biology or chemistry.
The researcher I knew was definitely not an engineer, so I assume he was on a lower payscale. Biomedical engineers typically get paid a bit better than the researchers everywhere.
This is an interesting observation, particularly when considering the engineering vs. science distinction that you mentioned. The value traditionally produced by life scientists is less predictable than it is for engineers, due to the more unpredictable nature of the business. To some degree, Google has probably overshot in terms of estimating the likelihood of success (the 'Andy Grove fallacy'). They've hired a lot of ex-biotech people though, so that's surprising. Perhaps they just want to ensure they attract the best life scientists despite the lack of a track record in the industry.
It seems to be a little random if GLS wants you or not, but if they want you they really want you. From my own experience and that of friends working in science-based startups, Google salaries are a major impediment right now for hiring. We're still able to get great people, but it certainly slows things down.