For speedy street routing, the Open Source Routing Machine: http://project-osrm.org/ Like Google's proprietary routing engine, it leverages the scale of local and regional travel (via contraction hierarchies, arterial travel is suited for precomputed cacheing); result: the system can be tuned to give near instantaneous results. OSRM had been led by Dennis Luxen of Mapbox, though he just moved to Apple. The OP (Mapzen's Valhalla) appears to have a similar approach as OSRM. Good libraries for other scenarios exist; e.g., cycling and multimodal planning -- see: http://wiki.openstreetmap.org/wiki/Routing and OpenTripPlanner https://github.com/opentripplanner/OpenTripPlanner
There's an academic Thesis on the principles of contraction hierarchies that is worth a look if you're in this space. http://algo2.iti.kit.edu/documents/routeplanning/geisberger_... My favorite is actually a Master's thesis that steps through the process of using contraction hierarchies to build a routing engine (MoNav) on OpenStreetMap data. https://code.google.com/p/monav/downloads/detail?name=thesis...
For nuanced or complex problems, set up your objectives and constraints against a good solver: http://en.wikipedia.org/wiki/List_of_optimization_software I'm partial to Google's OR Tools https://github.com/google/or-tools (Apache License).