Adding Likelihoods depending on RooArgList order

I have a problem of adding different likelihoods. If I import 3 (or 2) RooRealNll-objects from different workspaces in different saved ROOT files and I add them together using

# python
NLL_list = RooArgList()

Combined_NLL = RooAddition('nllCombi','combined NLL', NLL_list)

the resulting combined likelihood curve depends on the order in which I add the NLLs (see attached screenshot and full python-testscript without workspace Objects).
Also, the absolute value of the combined NLL curve doesn’t make full sense, as it is higher then you would expect from the sum of the 3 (expect ~ -11000).
All in all I cannot trust the resulting likelihood curve at all. Contrary to an example in a RooFit talk ([]School of Statistics 2008[/url]) I had to give each NLL-object an individual name first, otherwise only the curve added first is added to itself.

Does anyone have a solution on how to reliably add likelihood functions in RooFit? (2.97 KB)
Comb NLL ID401 ID3 ID6.eps (17.2 KB)
Comb NLL ID6 ID401 ID3.eps (17 KB)
Comb NLL ID3 ID6 ID401.eps (17.1 KB)