OK I found the issue(s).
1. the chanle sampling the DUT input had the 20MHZ bandwith limiter enabled. Disabling that, the result, at least with the power splitter, was as expected. One can argue if that is a bug in BODE II, as Bode II controlls the inpus it should be aware of the bandwith limiter in conflict with the sweep span.
2. Without limiter but spliting with a tee only, the plot looked nearly as wrong as before. Adding a further tee with a 50Ohm terminator, made things looking good. Well, with using probes, this issue wil not happen.