"Simply not worth the cost just to avoid drawing a few off-screen triangles" - THAT
I do not agree. You cant say that because you may be sending 100% of the model when you are actually seeing 5%. If the calculations per vertex are expensive or you are GPU bound (saying that cause my GPU is not the greatest and doing extra stuff seems to be unnecessary
It's not very good for overall conversation that you split quote like that.
Actual quote should be:
Subdividing meshes at run-time is overkill and simply not worth the cost just to avoid drawing a few off-screen triangles.
Which has totally different meaning.
But back to the point. There was great presentation at siggraph from Ubisoft. http://advances.realtimerendering.com/s2015/aaltonenhaar_siggraph2015_combined_final_footer_220dpi.pdf
They are doing super precise frustum and occlusion culling on gpu. Using just 64 triangle strips as primitives. But the conclusion was bit disappointing because they didn't get huge overall gains even after all that effort.