I think the other significant aspect is the speed of iteration. In my experience, producing good visualisations takes lots of iteration and those can take a lot of time if you have to alter each data item by hand. I've often reverted to paper and sketched out a revision that's way a few times before going back to the drawing program. I think it's likely that it's much quicker to iterate using D3.js (in the right hands).
As Bostock's incredible skill is deployed on a variety of visualization problems, the New York Times builds a library of visualizations and templates that they can use for the foreseeable future, as long as the data is in the same format. That is incredibly valuable.
It is starting out as artisan tool for programmers to build data visualization. Overtime, we'll have reusable templates/components that speed up the creation process and reduce cost. Christophe's list is awesome. I reference it quite a bit :)
At some point, creating vector graphic will be just as fast as creating static graphic. Plus, much more flexibility.
My team is tinkering with such idea with http://vida.io. Here's a little demo video:
http://www.nytimes.com/interactive/2013/01/02/us/chicago-kil...
I wonder if making static charts via D3js has some time-savings/production advantages when the dataset is large enough? Before you say "maybe they just wanted vector graphics that worked for high-res"...that's obviously a benefit, but not enough on its own to give the web devs a graphic that could've been done via the traditional means (many of the Times stories include static graphics as PNGs in the sidebar)