Hi Gerald, 
(I am responding in English so that everybody interested can follow our discussion.)
Many thanks for your question. For the SEMs we used the community weighted means calculated for each island (trait values from individual species were not included here). As we used 28 islands the dataset was sufficiently large to get robust model output.
But I think this is a good question whether and how the results would respond if using lower number of islands. We did not test this specifically but I would assume that (as with all models) you need a large enough number of data points to have enough degrees of freedom for model estimation. I used simple linear assumption here and thus only needed to estimate few parameters. This will probably change if you use other underlying models in the SEMs.
Regarding the number of species: this is important when estimating CWMs. We only included islands with at least five species present to avoid high stochasticity in CWM values on very small islands with only one or two species present (our dataset also included empty islands and many with only one or two species; you can check out the data here: Schrader, J., et al. (2019). Plants on small islands revisited: the effects of spatial scale and habitat quality on the species-area relationship. Ecography, 42, 1405–1414.).
I hope that helps. Many thanks again!
Cheers,
Julian