I disagree that the LN paper made conservative assumptions.

A "black swan" event like a major lightning hub going down may force more channel closings than the network (as currently implemented) has time to process before time-out.

The LN simply does not work if the base layer is congested.