I ran the Decoy Peptide Validation tool on both of the file you sent, using DECOY1 as decoy (ignoring DECOY0 that was used by PeptideProphet), the decoy rate is applied is 1/3 (DECOY1 are about 1/3 of the fasta database when DECOY0 are ignored.) These are the results I got:
As you can see from your decoy counts and from the above images, the new TPP is doing a much better job of controlling the error rate (as compared to an independent decoy set) on this dataset.
If you want to generate these images yourself you can use the tool in Petunia menu under "TPP Tools"->"Decoy Tools"->"Decoy Peptide Validation"
Let me know if you have any questions.