Adobe has a funny way of hilighting words. You have to tell acrobat the page and word numbers to hilight even there are no "word" boundaries in a PDF file. Occasionally special formatting within the file will throw off the appliance's idea of word number vs. acrobat's.
If you could supply a few sample files and queries and where the hilight is off we could investigate if there's anything that can be done to improve the situation.
Using acrobat reader 8 those searches in those files hilight correctly for me. However, if I use the "remove common" feature the hilights are off. "Remove common" and "keep tags" and "ignore tags" are not usable if you want accurate pdf hilighting. All of the text must be kept so word counts are correct.
Accuracy seems to be much better on pages 2+, but page one is still off...
It is better with "remove common" off, but it will remain so, I suppose, because we need to use "keep tags" to remove navigation links from search result abstracts.
The word counting resyncs at each page boundary so remove common should generally only affect the first page.
Keep/Ignore tags are far less likely to cause a problem because the tags you specify for HTML are highly unlikely to match anything in a PDF document. Remove common does not need to be on to use Keep/Ignore Tags.