Hi DrJ,
my feeling is that the time is not lost in the index processor but in the SODA processor that runs after it.
Results are possibly a lot better, if the number of results are smaller that come out of the index elimination.
Are you testing with actual real data or with a simulation?
Sometimes simulations tend to have huge resultsets that don't happen in practice.
We will review the SODA processor nearterm and that will very likely improve your results quite a lot.