It is because you don't compare the elements in a consistent way. For example,
s = struct;
s(1).x = 1;
s(1).name = 'John';
s(2).name = 'Sarah'; % Sarah has no x values because of some unnoticed bug
s(3).x = 3;
s(3).name = 'Robert';
[s.x].'
so you compare two element vector to a scalar (3).
The best way to avoid these, as you called it dangerous situations, is to make sure you don't have empty elements.
So correct way to implement this would be (one of many ways):
s = struct;
s(1).x = 1;
s(1).name = 'John';
s(2).name = 'Sarah'; % Sarah has no x values because of some unnoticed bug
s(3).x = 3;
s(3).name = 'Robert';
% Find empty elements and replace them with Nan
ind = cellfun(@isempty,{s(:).x});
s(ind).x = NaN;
answer = s([s.x].' == 3 ).name % The correct answer should be Robert
