Well, Bitcoin was designed 6 years ago by a mysterious individual and before there have been any such similar system. Since then there have been so many progressions, new ideas and new coins, and we now have so much knowledge that Satoshi hadnt while designing the original protocol. So I dont think that any arbitrary choice made by Satoshi is neither the only possibility nor the best one
What the authors are basically saying is that there is no need to make the average interval so long that there will almost be no natural occurring forks, meaning this interval doesnt have to be much bigger than the network delay. So you may think that 1block per second is too optimistic, but you must agree with the authors that GHOST can get the interval significantly shorter
Indeed I agree with you that this novel idea should be checked in reality before we make any change to Bitcoin (perhaps by building a new crypto-coin), yet I am very optimistic about that! The authors based their claims on
empirical results of the Bitcoin network and their work seems very reasonably to me (not like many other academic paper that have some unrealistic claims).
Lear