>>The behaviour observed does PARTIALLY make sense - why bother to change the table if there were no deleted records removed!?!
>
>But that would require additional work for VFP, to analyze if there are deleted records in the first place. Assuming that there will usually be deleted records, it seems much more efficient to do the PACK inconditionally.

It doesn't seem to be "additional" work... it appears to copy the table's undeleted records as per normal, then decides to quit IF it didn't encouner a deleted record along the way. So it just deletes the copy in that case (if it even wrote it in the first place (it may have fit in RAM)). Then it of course skips the REINDEX, but continues with the FPT read/analyze.