>Sergey,
>
>>It's actualy opposite. IN subquery runs once and than result is used by the main query. The EXISTS uses correlated subquery so it has to run for every record in the main query.
>
>This test code would seem to say that IN is doing a lot more work than EXISTS.
>
That's because you put f2() function in the wrong place in the EXISTS subquery. See revised code below.

CLEAR
create cursor x1 ( i1 i, c1 c(10) )
for i = 1 to 10000
   insert into x1 values ( i, transform( i ) )
endfor
index on i1 tag i1

create cursor x2 ( i1 i, c1 c(10) )
for i = 1 to 10000 step 2
   insert into x2 values ( i, transform( i ) )
endfor
index on i1 tag i1

n1 = 0
n2 = 0

select *, f1() ;
   from x1 where exists ( select i1  from x2 WHERE f2() AND x2.i1 = x1.i1 ) ;
   into cursor x3

? n1, n2

n1 = 0
n2 = 0

select *, f1() ;
   from x1 where i1 in ( select f3(i1) from x2 ) ;
   into cursor x3

? n1, n2

function f1()
n1 = n1 + 1
return

function f2()
n2 = n2 + 1
RETURN

function f3(tnVal)
n2 = n2 + 1
RETURN tnVal

However, this test only shows how many records where processed by query and subquery. I doesn't take in account other factors. If you run speed tests you'll find out that in VFP IN subquery is faster in most cases than EXISTS.