Sure, the molecular clock is always going to be inaccurate to some
degree--it's an estimate with error. Lots of useful scientific
estimates have this property.
For a particular case in which you want to know how severe the error is,
you could try a statistical test proposed by Felsenstein. Estimate the
phylogeny by maximum likelihood twice, once imposing a molecular clock
requirement, once not imposing it (programs DNAMLK and DNAML from PHYLIP
will do this). Make an appropriate likelihood ratio test to see whether
relaxing the clock assumption significantly improves the match between
the data and the best phylogeny. If not, it seems reasonable to assume
that the clock assumption is okay for these data.
If the non-clock tree *is* significantly better, comparing the two may
provide some insight into why the clock assumption is failing--I
recall a case, for example, in which one of the loci examined turned out
to be a pseudogene copy (which was, of course, evolving much faster than
the others). This was obvious in the non-clock tree as the branch
leading to that species was greatly elongated.
Mary Kuhner mkkuhner at genetics.washington.edu