Some physicists in Korea and New Mexico have been making computer models of road use. They say that, counterintuitively, we've got a bunch of streets that actually cause congestion, and removing them (e.g. turning them into greenways) would reduce congestion on surrounding streets.
From this map (helpfully extracted in this post by The New Republic's Bradford Plummer) it looks like the dotted black lines are on West Houston Street, West Broadway, Kenmare Street, Grand Street west of Lafayette, and Canal Street west of Sixth Avenue. These lines show that if you made the streets unavailable to cars, congestion would improve on the rest of the grid.
This is due to something called Braess's Paradox, based on models worked out by John Nash. The paper (PDF) by Youn, Jeong and Gastner was discussed in a Christian Science Monitor article and WorldChanging; I found it by a brief link from The Overhead Wire.
Their model seems to be based only on trips from Washington Market Park in Tribeca to the Queensboro Bridge. It's not clear to me whether it would hold if you take into account all the other origin/destination pairs.