We've accidentally conducted the experiment multiple times-- heading from northern Marin into SF with visitors in another car and failed to tell them to ignore the google 'faster route' -- they arrive much later after detouring through the east bay and getting stuck in traffic coming down to the bay bridge or inside SF.
I have no idea how Maps estimation works, but it's easy to imagine that it could make a decision based on a brief delay that will likely have cleared before you arrived, while the path it suggested congested ahead of you.
Even if they even managed to accurately give a solution that was correct in the expectation, a much longer route is almost always going to have a much higher 95th percentile time (unless the shorter route has a train or drawbridge or similar).
If given a choice between of an alternative that would be 1 minute faster on average but never more than 2 minutes faster and 5% of the time you're advised to use it it's actually >30 minutes slower-- which would you take?
But I doubt they're particularly accurate in terms of the expectation in any case-- at least not relative to traffic driven 1 minute differences.
Aside, I drive a stick, so I don't especially love stop and go-- but Maps has done far worse to me in an effort to avoid some stop and go on an arterial road, e.g. sending me off onto extremely hilly residential streets clogged with other drivers, leaving me chirping the tires to avoid rolling into the bumpers behind me in the Maps congo line.
It's better to be in congestion on a road that was designed for it than to still be in congestion on some detour in an area where the roads weren't designed for heavy through traffic.