Take another look at ggplot2; it's exactly the "collection of composable primitives for building statistical graphs" that you mention; but it also gives you some higher level functions that give you a starting point that you can then customize.
There's a benefit to working in Hadley's / Wilkinson's grammar, but the tradeoff is that it obscures the underlying DOM representation (or R's graphics subsystem, as the case may be).
ggplot is for analysts, and c2 is for web devs and hackers.