In City Journal, Brandon Fuller writes: The Genius of Dirt Roads :
"Angel writes that governments in the developing world, whose financial capacity is often limited, should focus on what may sound unglamorous: establishing an arterial grid of dirt roads throughout each city’s future territory, much as the commissioners did in Manhattan. The grid should connect to the city’s existing network of roads, of course, and it should cover an area that the city expects its future population growth to require. These arteries will one day carry public transportation and private traffic, and such infrastructure as water mains, sewers, storm drains, and telecommunications networks will follow their routes."
The grid has advantages and drawbacks. In Planning for Place and Plexus we write:
The morphology and queuing properties of the plexus (its supply and demand) ultimately determine both the efficiency of the network in moving people and the efficiency of the land use. Radial (hub-and-spoke) networks allow easy access to the center but create inconvenient sharply angled parcels. In contrast, 90-degree grids maximize travel times (for anyone traveling in a diagonal direction) but create efficient parcels. A major issue with network topology is the interconnectedness of the network. Interconnected networks, be they grid or radial in nature, enable and even encourage through traffic, while a tree-like network discourages that problem. The topology of the network, grid, radial, organic (curvilinear) or otherwise, affects its performance.
The regular grid (with occasional interruptions) is arguably the most common topology for cities. It has been employed in cities for millennia. In the United States, the most influential legislation affecting the morphology of roads was the Land Ordinance of 1785. In many respects, it laid the foundation for future land use-transportation policy by adopting the Public Land Survey System, creating townships and subdividing them into 36 sections of one square mile (259 hectares) and 144 quarter-sections of 0.25 square mile (65 hectares) each. Roads delineating each of the sections were referred to as “section roads.” Subsequently, many urbanizing areas continued to use the centerlines of those roads as the location of present day arterials; the arterial networks are often further broken down into a finer grid of blocks.
A key point that has not been generally considered is the flexibility that the uniform and undifferentiated mesh networks (termed “grids” here) provide to changes in land use. A uniform grid allows alternative spacing between activities, spacing that can change with economies of scale. For instance, consider retailing. As described in Chapter 9, many stores—especially grocery stores—have been getting larger, while their numbers have dropped. Many New Urbanists, who advocate small-scale neighborhood retail, bemoan this phenomenon. Suppose that economies of scale indicate that it is efficient for the average retail store of a certain kind to increase in size from 1,000 to 2,000 ft2 (93 to 186 m2). Previously there may have been one such store every 10 blocks (one for every 100 square blocks); now there can be one every 14 blocks (one for every 200 square blocks). A grid allows the flexibility for re- spacing while keeping nearly optimal size stores. ...
A tree network, in contrast, fails to provide such flexibility; a store can locate either at the neighborhood center, at the community center, or at the regional center; it can serve perhaps 5,000 people, 15,000 people, or 60,000 people. A store optimally sized to serve 10,000 people cannot be located at a consistent node level—or, if it is, it cannot be efficient. A firm may need to locate stores in some neighborhood centers and not others, causing people to go into other neighborhoods in some places.
Recognizing that grid-based road networks might not lend themselves to locations that were not situated on flat, featureless plains, designers introduced several variations. To conform to the contours of the land, Frederick Law Olmstead employed curving streets in many of his designs (e.g. Roland Park, Maryland). Permutations continued to evolve over the years, and the “loop” and “lollipop” designs became the standard in suburban settings
I think the idea of a particular network topology (grid vs. tree) depends a lot on the topography. Getting this right is important. The idea of laying something out in advance (Angel's main point), so that property rights and development can occur on that lattice, seems a very good one.