Percentage quantifiers (%Qs) such as *thirty percent* have been shown to give rise to conservative (C) readings (*The company employs 30% of the women*) as well as non-conservative (NC) readings (*The company employs 30% women*). In languages like English, the NC reading is unavailable in subject position (*30% women work here), which has been labeled 'subject-object asymmetry' (SOA), whereas in languages like German the NC reading is also possible with 'subjects'. We show that both C and NC readings also exist in languages that do not display a morphosyntactic distinction between the corresponding %Q constructions, namely Slavic languages without articles. Based on data from corpora and cross-linguistic questionnaires, we make the novel empirical generalization that word order plays a crucial role in distinguishing between the two readings, irrespective of whether a language additionally marks the difference between the two by the use of definite vs. bare nominals (German, Bulgarian and Macedonian) or not (the other Slavic languages), and that this also accounts for the SOA: Languages with a rigid word order (e.g. English) do not allow for NC subjects, because subjects necessarily have to appear sentence-initially, whereas languages with 'free' word order (German, Slavic) do, because subjects can stay within the VP. We argue against previous accounts that ascribe a crucial role to focus for the NC reading to arise, in taking focus to merely be derivative from the requirement of NC %Qs to appear low, paired with a general rule for sentential stress placement.