Joost's advice about using Perl 5.8.0 is on the mark,. even if your HTML input data is not unicode -- and it's very likely that your data is something other than unicode, such as ShiftJIS or god-knows-what (I hope you know which encoding you are dealing with).

Not only are perl-5.8.0's strings stored as utf8 internally, but the Encode module, which is part of the 5.8.0 distribution, provides the means for converting back and forth between utf8 and a wide assortment of alternate character sets, including all the major (pre-unicode) Japanese standards, as well as the other forms of unicode (i.e. utf16, big- or little-endian).

And the new tricks that you get to do with regex matches, involving predefined unicode character classes, are truly awesome. Not only do you avoid nefarious corruptions of multi-byte characters completely, but you get to match characters according to what they really are.