Whether or not ClustalW decides that it is 'good' to have a gap in an alignment depends of course on the gap penalty. For very low gap penalties, ClustalW may easily insert a gap to get a 'better' alignment, even if a deletion or insertion would be unlikely (when could this be the case? Think of the relation between nucleotides and proteins). For extremely high gap penalties almost all gaps will go away (except at the beginning and end of the sequence), even if a deletion or insertion could easily have taken place.