A question on VECM

Hi all, in Lutkepohl, page 250, I found that if there are stationary
variables in integrated system then they must be put in upper r-dimension.
My question is, is that the fact? If I do not do it, is there any problem
in estimation and interpretation? I have done few exercises and found that
parameter estimation is not infected with ordering (except IRF
estimation). Even is page 303 an example is presented wherein inflation
rate variable is taken as 2nd variable, although it looks like a stable
process. 

Can anyone please clarify that? If really ordering is a problem in
presence of a stationary variable, can anyone provide me an example, with
perhaps in R-code so that I can regenerate? 

Thanks