Regularization in parameter estimation

I'm estimating a nonlinear grey-box model with pem (estimator 'lsqnonlin'). I would like to add a regularization factor to the optimization, to penalize high-variance estimates.

I tried, following http://www.mathworks.it/it/help/ident/ref/idnlgrey.html , to set up the L2 regularization in this way, being nlgr my idnlgrey object:

nlgr.Algorithm.Regularization.Lambda = 10;

I didn't explicitly set the weighting matrix for Lambda, as from the documentation it should transparently defaults as mat(1) (which is what i want). I also tried other (higher) values for Lambda, but apparently the results are identical with or without this setting, and I obtain very high varianced estimates.

Am I missing something?

Note: I am using Matlab v. 2013a

[maru]

apparently, it seems to completely ignore the configuration.