The quick and dirty way would be to do a second, non-greedy replace, restoring the content of the tags - something like this (untested):

s/<a (.+?)$__highlight_to_switch(.+?)>/<a $1$in{highlight_text}$2>/gi
[download]

The not-so-dirty way is to use one of the HTML parsing modules, and replace only appropriate parts of the page in the first place.