Cohadar once said something (within the first two pages of the TT thread off memory) about evaluating return-nothing boolexprs causing desynchs. So unless that is tested I don't consider it safe.
Then again, return-nothing and null are two different things. So I suppose it may be fine. I'd still want to see it tested, though.
(Obviously people do null boolexprs all the time. Problem is in evaluating a destroyed boolexpr, that handle may be reused, and you may try to evaluate a unit or something dumb or a newer boolexpr, which may execute code you didn't want to execute.)