No expert on this - are you sure the 33uF caps are ok?
In what way, as in are they damaged? I doubt that. There are two series strings of 13 caps. I had one initially (18uF) and I added one more to lower the ripple. Now the ripple is fine (150V at most - with a 330mH choke at output side), but the DC drop is there.
My understanding always was this is normal. I always expected the no load DC to be 1.41*AC(RMS) and at load to drop roughly to AC(Rms), but then I was told about those other PSs that don't drop as much.
I've managed to simulate the entire circuit in ltspice to test few things. I've measured the transformers parameters (winding inductance, series resistance) and I adjusted the coupling factor for the simulation no load secondary voltage and primary current to match.
Ltspice shows the same dc voltages I'm getting in reality. Unfortunately I can't reliably measure AC at 2.3kV so I have to go on what LT spice tells me. Based on that there is barely any voltage drop on the transformers (50V or so).
The capacitors discharge from peak, and on the next peak they don't charge fully. It is as if something is preventing higher current from flowing. I'm guessing transformer'secondary inductance, connection lead resistance, capacitors ESR. Connection leads are at most few tens of milliohms, the inductance of primaries is pretty large (15H each, 4 in parallel are 3.75H), also the resistance of secondaries is pretty large at 140ohms (35 across 4). Capacitors are electrolytic caps (13 in series). If everyone od them is 10 ohms, they contribute 100 ohms total.
Charging currents are slightly under 2A per transformer.
People that have those power supplies that stay close to AC peak at their rated current all seem to use oil filled capacitors rather than electrolytes, but I'm not sure this is relevant, because my ltspice simulation uses ideal caps (no esr) and still there is same voltage drop. I'm leaning towards blaming transformer inductance, or diode overshoot.
So I'm thinking, perhaps if I used fast, or ultrafast rectifiers, that would help? Typical recovery time of normal rectifier diodes is on the order of 1uS. One cycle at 50Hz is 20mS. If the charging time is 5% it would have 1mS to charge. It seems a very long time in comparison with 1uS, but then there is overshoot that normal rectifiers have (they continue conducting for a brief period after the AC waveform crosses zero allowing reverse current flow). Ultra fast rectifiers don't have overshoot allegedly.
I can add more transformers to deal with a nductance, but I wonder if there is anything else anyone can suggest (othen than buying a bigger transformer)
Edit: also the problem with 3 phase is that no load DC voltage would be very high - with ac of 3500V it would be almost 5kV. I would need to waste half of the power of the PS in a bleeder resistor to bring that voltage under 4kV (max tube voltage). A large choke in front of the rectifiers would bring the voltage down, but it would cause all sorts of problems at rapid changes of load (ringing, voltage spikes etc)