Thanks Sergey,

we are to the usuals:
it must try one hundred thousand sequences of code in order to understand as this works....

The problem is that if is full optimizable at the moment of the SELECT,
VFP creates a filtered cursor,
if subsequently this filter becomes partial optimizable, the filtered cursor can become extremely slow.
If the things were documented better, Thread ID of UT would be < 500000.

Fabio