I've seen this sort of behaviour before as a result of the "AR" (auto regression) filter which seems to auto-enable itself in some situations/scenarios...
[I haven't identified the pattern yet, maybe it's related to the template used? Either way, whenever I come across screwy results with frequency domain 'ringing' like yours, I go hunting for it in the simulation settings dialog]
The filter intends to speed things up by truncating the time-domain simulation and inferring the corresponding frequency domain. I'm not sure when it's meant to be useful (or what I'm doing wrong) as it always seems to give me results like yours. Turn it off (if it's enabled) and try again