For example, a route from SOMA to Nob Hill can be made flatter by approaching from either the east or west. Technically the "flattest" as measured by elevation gain between the two is straight up Taylor from Market, but it's much nicer to bike to the top of Nob Hill by going out of the way (e.g. Polk → California, Embarcadero → Broadway, etc.)
Most people on most bikes can only take so much grade before climbing becomes impossible. A maximum grade setting would be very helpful for that exact reason.
My reasoning for the bike-hike being faster is that legs are designed for it. Legs are only inefficient on flatter ground because you have no equivalent of high gear (like extendible legs), so all that torque is going to waste.
Legs have to start, go up, go down, stop.
Wheels go brrrrr.
And almost any practical routing system will have a mechanism for tradeoffs of different aspects. Often there is a distance/time tradeoff which directly relates to speed on corresponding segments. But the tradeoff can also be in the form of "on average x seconds spent waiting green light in each intersection or time to stop before railroad crossing. So there is no reason for a system optimizing grade to completely ignore all other factors, it's just a question of weights and curves of each of them.
your example (from market & taylor to the top of nob hill) is a nice illustration: (1) straight up taylor is 229 ft of climbing at 25%, (2) via polk and california it's 251 ft at 13%.
as you probably noticed, we only look at the total elevation gain, so taylor "wins"... but I see the argument for preferring an extra 20ft of a more gradual climb.