Looking at that document and the criteria used, I'd say the narrower definition (giving 3%) is a better measure. The 12% figures includes many questionable criteria - including cases where potential victims were intoxicated. It also appears that it is a retrospective assessment - there's no indication that these people were cautioned let alone prosecuted.
Many rapes go unreported, where (for a variety of reasons) a formal allegation is never logged. Women are acutely aware of the perils of making such an accusation, let alone a false one. Obviously we will never know for sure but there is no doubt that rape is an underreported crime. As such, if we were to look at proven false allegations as a proportion of potential cases then I suspect that % would reduce considerably.