Word Index Performance

Openedge 11.5   Windows Client  Unix DB Server

We are use a word index for some tables in our system.  We allow the users to put their own keywords in a field on the record, then we append our own internal system word containing multiple key fields from the record.  An example value would be: "white shingle ridge @SysId%0501@Act%Y@Stk%Y".  The table has > 4 million records, when a query is performed on the table using the word index, I get the following results:

With a search value of "white* @SysId%0501@Act@Y@Stk%Y" the query takes 22.4 seconds returns 138 records and has 3,508,392 index reads.  

With a search value of "white* @SysId%0501@Act%Y@Stk%Y*" (* on the end of our internal system word), the query takes 2.9 seconds returns 138 records and has 3,526,291 index reads.

It seems counter-intuitive that the query would run faster with more index reads, can anyone explain what is going on here?

Example query:   for each item_branch no-lock where

                                                item_branch.keyword_string contains "white* @SysId%0501@Act%Y@Stk%Y" : 

                                end.

Thx, Blake