It's pretty wild seeing how off the estimates are and how far things have come since then. This page about limits lists the expected fastest average at 11 seconds and the best possible average at 5 seconds [1].
Today, the best solvers have gotten multiple sub 5 averages [2] and most good solves can get sub 11 second averages with practice. In fact most intermediate solvers can get sub 17 second averages with advanced cube technology (better turning speed and fewer lock ups). I've even seen a 2.35 second solve [3]
[1] http://www.ws.binghamton.edu/fridrich/hints.html#limits
[2] https://www.worldcubeassociation.org/results/rankings/333/av...