I'm putting together an 807 push-pull amp. Desired power out is 25W or
so.

Plate supply (5AR4, choke-input) is about 370V-390V as current drawn
goes from 150mA to 90mA. It goes up to 470V with no load. The
transformer and choke are rated at 150mA but I've run this supply at
150mA for a whole day and it's just getting lukewarm, so I think I can
draw a little more on peaks without any big problems.

Grid bias (-20V or -25V) will come from a separate transformer
winding/pot.

For the screen supply, it looks like my options and
advantages/disadvantages a

0. Ultralinear. Won't work because plate voltage is much higher than
rated rated screen voltage (300V). This is where an 807 differs from,
say, a 6L6GC type it seems. Elsewhere in some 807 data sheets they rate
triode-mode operation (plate strapped to screen) to 400V, can this be
extrapolated to saying that it's OK to run from ultralinear taps at
400V? How about at the 470V that my power supply puts out with no load?

1. Dropping resistor from B+. Not good, because variation with plate
voltage and screen current drives up distortion considerably.

2. A stiffish voltage divider from B+ supply. Appears to not be
particularly stiff (screen current will probably go back and forth from
5 to 20mA depending on signal) although some electrolytics can probably
clean this up. But "stif" here means probably 60mA (three times the
peak screen current) through the divider chain and that's eating up a
lot of my power transformer's capacity. Is 60mA too stiff, not stiff
enough, ???

3. Separate transformer winding/rectifier/filter. Could do it if I had
such a transformer and filter, but no more space on the chassis...

4. Couple of voltage regulator tubes in series (3xVR-90's? VR-150 and
VR-105?) with a dropping resistor to B+. VR tubes seem to be cheap and
plentiful and are even kinda pretty glowing.

#4 looks like like the way to go. The delta in screen current I read
from the 807 curves make it look like 5-20mA is the right amount for
two tubes, and this just happens to be the range over which some VR
tubes hit the application exactly.

Are there any gotchas with using VR tubes here? If one of the VR tubes
fails to fire then they don't regulate at all, but a good choice of
resistor to B+ will limit the screen dissipation to a survivable amount
if this happens, right? Or am I looking for a complete lightning storm
if a VR tube doesn't light?

I don't see a lot of other designs using VR tubes for screen supply in
audio amps, although in my long experience as a ham this is very often
seen in RF stages.

Are there gotchas with power-on voltage sequencing that my choice of
tube rectifier (5AR4) is going to cause?

The B+ current drain goes up by another 25 or 30mA or so by getting
screen voltage from B+, but I think I can handle that much more current
drain there.

Did I miss the boat completely in terms of guesstimating required
divider stiffness in #3?

Tim.