Fabio,
There is no real way to calculate it from VFP's perspective. Since the CDX uses run-length encoding in it's compression of the keys the size is dependent on the the actual data in the table.
I think the best you can do is to track the DBF vs CDX size for each table when the CDX is first created and then monitor the changes and reindex when a threshold limit is exceeded.
>With PACK ( it copy the DBF,FPT and recreate CDX) the solution is simple, but i want use this shema: