I had never heard the ceramic plate caps could cause such additional capacitance, seems counter intuitive.
Rich
The motivation for the statement is due to surrounding the bare metal of the plate cap with ceramic whose dielectric constant is about 10. So it is not counter intuitive, however it does beg the question, where are the set of metal plates creating such a shunt C on the plate. One is the plate itself, but where is the 2nd plate. Again, I suspect this is not the issue in your case, however if you are trying to chase stray C you begin to question all possibilities. The original article looks straight forward. Rich, you may want to reduce your troubleshooting to the most basic circuit elements. Hence, just consider the PI network only terminated in the total Rp of the tubes alone, 4K I believe. And see if that is straight forward to match to 50 ohms. For a Q of 12-15, it is easy to determine the presetting of the two C's at 21.3MHz. Then attempt to find a tap point on the PI net inductor... if possible, that would be L1 I think.