Text index statistics

I'm investigating why the optimzer is choosing different plans on a query involving a Text indexed table being joined to a non-Text indexed table.  I ran an event 10053 trace on a query with CONTAINS values that perform and the same query with CONTAINS values that don't (both are simple expressions like "CONTAINS(column,'(word1)'").  I can see that the optimizer is calling ODCI functions to get cost information from Text.  In the query that performs well, a cost value of 500,000 comes in and the join uses an index on the other table in the join, while for the badly performing