The results will vary based on the environment and topology of an area, not to mention the capabilities vs load on a particular graphics card.
Smaller terrain chunks allow for (under some circumstances) more optimal culling to be performed with a trade off in additional draw calls and higher overhead. Larger heightmaps require fewer draw calls but are more likely to not be culled since a portion of the heightmap is often visible increasing the (a different type of) load on the renderer.
As for what is more optimal for a given area, the only way to know is experiment with the tradeoffs to see what is best.