Just a guess here...

But as your adding (INSERTing) truckloads of records, VFP is constantly updating the internal tree structure of the keys in the CDX. After INSERTing tens of thousands of records, it may feel it has to rebalance the tree and it may rebuild it at the end of the file and just leave the original tree structure intact (but unavailable) at the beginning of the CDX, therefore taking up space.

I'll bet the REINDEX (and also INDEX ON) are more efficient since they both know how many records there are to index up front and therefore can make a better prediction as to how to structure the tree up front; therefore, the final CDX is "tighter".

Of course, I may be completely wrong... it's just an educated guess. It's been a looooonnnng time since I've studied CDX behavior.

--Brad