The confusion of the two labial letters is one of the standing textual problems of Indian and Buddhist manuscript studies, and it runs across scripts rather than being particular to Siddham. It has two independent causes and they reinforce each other. The written shapes are close in many hands, so a copyist reading from an exemplar may misread. And in a number of the spoken languages of northern and eastern India the two sounds merged, so a copyist working from dictation, or writing a word he knew from speech, may write either.
Because the causes are independent, the error can enter at any point in a chain of copies and can enter more than once. Manuscripts of the same work regularly differ on exactly these letters, and the variant is so common that it is often not even reported in an apparatus.
For a dharani or a mantra the stakes are different from those in prose. A doctrinal sentence usually survives a wrong letter, because context restores it. A string of syllables transmitted for its sound has no context to restore it with, and a variant there cannot be resolved from sense. That is one of the reasons the East Asian tradition treated the Siddham original as worth keeping alongside the Chinese transcription: two independent witnesses to a sound are better than one, and they fail in different ways.