Manuel,
Note that the models 1, 2 and 3 are not repeated measure models. (You
probably are aware of that)
I think but I would have to check back that the models 4 and 5 are.
I think that giving a correlation structure for model 4 would help the
estimation of the random effects. Now R is estimating the correlation
structure itself and this is laborious. PdDiag are normally good
structures to start with but I would try others ones.
Frank