How does increasing the number of samples increase quality of the voice generated?

[DamienMcE]

I've been experimented with trying to clone British actors Stephen Fry and Hugh Laurie. I've noticed that Stephen's voice is excellent for most common words but goes of the rails entirely if I say something too unusual (Stewart confuses it). Been using 10 x 10 second samples but have hours of clean audio that I could use. My questions are :

• Would I get dramatically better results if I increased the samples used to say 100? Or are there diminishing returns, as I add additional samples.
• Does the inclusion of a specific word in the samples improve its pronunciation in the final cloned voice? Should I be aiming for a large variety?