My idea for variable movement costs was to replace edge costs with enums of data about the tile and just switch depending on the moving pawn. I have no idea how that will impact performance but your method of separate functions might help with that too.
For multiple tile units does that array save much time even if you need to check the same tiles for something else? For example if a 3x3 tile wide unit moves forward even if your array says the next space is walkable for a pawn that size you still need to check the 3 tiles it is moving into for rough terrain, height, ect. or have an array for attribute that affects movement.