Walter,
Are you sure the performance problems you are seeing with SQL is not just an issue of the SQL optimizer making incorrect judgements about what to join first? Your xbase code uses a priori knowledge of your data when you set it up. If you have a select with a three table join in VFP you can quite often split it into two selects that perform much better using the same a priori knowledge to reduce the amount of rows to be joined.
>I've ran another series of tests here and I come to the conclusion that performance gains can be HUGE to even a factor hundred. With the test described in the thread 'Differences between xBase and SQL implementations' i've managed to see differences where the performance gain was more 1700% native: 1.2 sec against about 21 sec SQL.