> I guess I don't see why that would be the case.  He doesn't
> actually change the RTO estimation algorithm, so I don't see how
> his code would make things any _worse_. 

Right, the algorithm is not changed.  However, if you look at our
SIGCOMM paper the minimum RTO and the granularity of the clock have
an impact on the performance of the estimator even without changing
the algorithm.  We found that increasing the aggressivness of the
RTO timer trades off spurious retransmits (which hurt performance,
as well) in real networks.  So, you may end up hurting performance
by reducing the granularity.

allman