nlme and NONMEM

I'd appreciate hearing from anyone (off list if you think it more  
appropriate) who can share their comparative experiences of non- 
linear mixed effects modelling with both nlme and NONMEM. The latter  
appears the traditional tool of choice particularly in pharmacology.  
Having built up some familiarity with nlme I am now collaborating (on  
a non-pharmacological project) with someone strongly encouraging me  
to move to NONMEM, although that clearly represents another  
considerable learning curve. The main argument in favour is the  
relative difficulty I have had in getting convergence with nlme  
models in my relatively sparse datasets particularly when (as in my  
case) I am interested in the random effects covariance matrix and  
wish to avoid having to coerce it using pdDiag().

I note the following comment from Douglas Bates on the R-help archive

The nonlinear optimization codes used by S-PLUS and R are different.
There are advantages to the code used in R relative to the code used
in S-PLUS but there are also disadvantages. One of the disadvantages
is that the code in R will try very large steps during its initial
exploration phase then it gets trapped in remote regions of the
parameter space. For nlme this means that the estimate of the
variance-covariance matrix of the random effects becomes singular.

Recent versions of the nlme library for R have a subdirectory called
scripts that contains R scripts for the examples from each of the
chapters in our book. If you check them you will see that not all of
the nonlinear examples work in the R version of nlme. We plan to
modify the choice of starting estimates and the internal algorithms to
improve this but it is a long and laborious process. I ask for your
patience.

Can Doug or anyone comment on whether the development work on  
lme4:::nlmer has included any steps in this direction or not?

Thanks

Rob Forsyth