>I'm curious about the performance of the ADO intermediary that Alexander suggested.

This is test code:

function BigXMLTest( n )
oOleDb = CreateObject("OleDbFox.Utils")
oRecordSet = CreateObject("ADODB.Recordset")

close databases all
Erase "x1.xml"
Erase "tmp.xml"

create cursor x1 ( i1 i, c1 c(10), t1 t, i2 i )
for i = 1 to n
   insert into x1 values ( i, str(i), datetime(), int( rand() * 100000 ) )
endfor
? n

i = seconds()
cursortoxml( "x1", "x1.xml", 1, 0 + 512, 0, "1" )
?? seconds() - i

i = seconds()
oOleDb.CursorToRs(oRecordSet)
oRecordset.Save("tmp.xml", 1)
oRecordset.Close()
?? seconds() - i

close data all

i = seconds()
xmltocursor( "x1.xml", "x2", 512 )
?? seconds() - i

i = seconds()
oRecordset.Open("tmp.xml")
oOleDb.RsToCursor(oRecordSet, "x3")
oRecordset.Close()
?? seconds() - i

My results (on P3-600):

    n CursorToXml  XmlToCursor
        VFP   ADO    VFP   ADO
 1000 0.040 0.130  0.491 0.160
 2000 0.070 0.270  1.572 0.281
 4000 0.150 0.521  5.808 0.581
10000 0.371 1.412 92.513 1.472

Not very bad, isn't it?