I finally did some decent bench testing on the EFB mod using my diytube ST35, a stock Rev C unit with mid 60's complement of transformers pulled from a SCA-35 years ago. I only had about half a day to dedicate to testing and modding, and focused on testing at 13.65W output, which is 1dB down from the 17.5W output (originally spec'd by Dynaco as never "exceeding 1% within 1 dB of 17.5 watts each channel"). This test was done on one and two channels over three amplifiers: a stock diytube ST35, then with all jumpered cathodes (like original), and then with the EFB mod (stock mod). The power tubes were two matched pairs of 6P14P. Voltages were adjusted to original specs with Variac. No tweaking took place. No IMD tests were done.
With all the testing, I realized I still liked the goofy individual biasing scheme of my original design. My 6BQ5/EL84 collection is a hodge podge, and the current design meets those needs. The difference - at least in these tests - just doesn't seem like enough to justify doing major layout changes. Note that the key data is with two channels being driven. So no Rev EFB at this time. I did start the layout changes, but with the individual biasing scheme I would have to invest too much time: probably several weeks of diytube time. My apologies. If I was doing this full-time, I would definitely do the mod.
I noticed a major performance hit from using some of my old US tubes with weaker transconductance (this seems obvious of course). So a new quad with stellar transconductance is the key to max output performance. The older tubes are handier for the guitarists. I think JJs would work particularly well with the EFB mod, as they have very high transconductance and bias hot which the EFB would handle. And the reduced bias requirements should keep those tubes running stronger longer. That could be a killer combo.
EFB mod tips: I used a Dremel tool with a cutting wheel for the screen & grid stoppers. I quickly was able to make eight precision cuts, and then carefully stripped the solder mask away from the traces with the same cutting wheel (scrape with Xacto if you don't trust your dexterity). I then used 1206 sized surface mount 1K and 100 ohm resistors. Don't worry about their voltage rating as the voltage drop isn't much. You can also use axials with trimmed leads. The entire cathode networks can be removed: R17-R28, R43-R46, C5-C8. I jumpered each side's cathodes at C5 & C6, and C7 & C8. Everything else can be put on perf board including the 5 ohm resistors. I used a clip-on heatsink for the LM337 which kept it cool. You can get LM337s on eBay for less than doing a full blown Digikey or Mouser order, too.
Results for One Channel (PDF)
Results for Two Channel (PDF)