About testing the relevance of SQL2000 query results

The sort sql2000 query results are sorted by relevance, which may sound appealing, but can really be achieved.
I saw an article in the morning on the Internet using Microsoft Index Server to do Full-text query (this has been seen before, in the computer Management also brought such a query function)
My IIS Default Web server has more than 100,000 HTML documents under G:/wwwroot
Test: Strsearch = "Select DocTitle, Path, FileName, characterization, Size,write,rank" & _
"From SCOPE ()" & _
"WHERE CONTAINS ('" & Request.Form ("Txtsearchfor") & "') Order by RANK Desc" also has a sort of correlation, I do not have the specific cost of time
Calculation, but the feeling is acceptable, when the page is very fast. The biggest drawback, however, seems to be that only static pages can be indexed.
In the afternoon, I indexed the database of a previous 50多万条 record (mainly song name and singer name) in sql2000, and I could start testing at night.
Test one: "SELECT top * from Song1 where contains (SongTitle, ' love ')", without any processing of the results, just according to the ascending sequence of IDs
The time cost is basically maintained at 0.016s, the speed is very satisfying, at least not feel slow.

   Test Two: The rank value is used to sort the correlation, and the result of the query is satisfactory in the quality of the order, both by rank desc "or" Orders by rank ASC ", whether the query uses or OR And do a lot of keyword sorting is OK, but the cost of time I can not bear, incredibly in 6s to 8s,
and CPU is also occupied by a higher
   I see other online search on the relevance of the ranking is relatively fast, Open source Lucene I have not studied, because I do not understand Java.
But I think if you query the correlation for each keyword at index