Quick diagnosis: In cases 2 and 4 the function SVAR_coint is exiting early on error, because jbeta has 5 rows and the function thinks it ought to have 6. So on exit the "SVARobj" has not been filled with the required matrices. Then when we get to the function vecm_est, jbeta is 0 x 0 and the attempt to extract a non-null submatrix necessarily fails ("Index value 1 is out of bounds").

I've taken the liberty of making SVAR_coint fail with an error message (in git) if jbeta is deemed not to be of the correct dimensions. Of course feel free to revert that, but it seems some more error-reporting is needed.