This post is kind of long because I want to post what I know at this point about this problem and what we are doing about it.
For the stock Paramount in 2A3 configuration (direct coupled), we have seen no problems with old stock 2A3s, or with the supplied Sovtek tubes. A few modern-production tubes have shown a visible and/or audible arc and "pop" through the speakers; this has been known since the ParaGlow days (predecessor to the 2A3 Paramount) and the usual solution was to implement a standby switch on the high voltage supply - or use a tube that does not have the problem. This is, incidentally, why the Paramount power supply circuit board has a place for that high-voltage switch - it's not mentioned in the manual where a jumper closes the circuit, but the jumper is in the circuit at a place where a normal 240vAC switch can be safely used to switch on the high voltage.
More recently we (well, actually a customer) found a severe problem with the EML 2A3 - the cathode emission is destroyed in a few startup cycles. Apparently it can be restored with an extended burn-in but nobody knows how much ultimate lifetime may have been lost. Jac at EML figured out a lot of this - credit to him for that work! This is so severe a problem that we have been working much harder on the "soft-start" board to see if we can make these highly-regarded tubes work with this circuit.
The issue, as far as I understand it, happens as the tubes and power supply start up. The direct-coupled 2A3 grid is at high voltage as the power supply ramps up (fairly quickly) and before the 2A3 filament comes up to temperature. I can speculate all day, but it's hard to prove exactly what happens next; I'll just say that the 2A3 cathode is not protected until it emits enough to form an electron cloud around itself. Sometime during that unprotected period there is an arc with some tubes. It probably depends on exactly how the active cathode surface is built, and on the speed with which it starts emitting. In any case, some tubes are tough with respect to this abuse, and some are intolerant of it.
Our approach right now is to delay the current source feeding the driver plate. That alone is not enough, the leakage current before the C4S stabilizes still causes a tiny "pop". We've added a resistor bypass from driver plate to ground (330K is the current value) which holds the 2A3 grid very close to ground until the 2A3 warms up - actually the delay is long enough that the driver also warms up. Then the driver high voltage ramps up slowly to its stable value.
(In the course of this work, I found that the driver plate voltage could vary a lot with different tubes, which can play havoc with the 2A3 operating point, power, and distortion. Making the driver plate voltage adjustable would not only solve that problem, it would also make it easier to swap in different tubes. Therefor this feature was added to the board - I am now at version 4.4 of the board design. :^) We tried a servo bias design but it did not sound quite as good as a fixed bias voltage, so there will be a trimpot on the new board to adjust the driver plate voltage to the optimum value.)
To fit all these functions in, without adding a whole new logic board, relays, and power supply, some compromises must be made. The biggest one is that, once the amp is turned off, it must remain off for about 30 minutes before it is turned on again, or else the delay time will be reduced and produce arcing again.
I am certain this will reduce and in most cases eliminate the "pop". I am not yet certain that the EMLs will be free from damage, though I think we have addressed the main cause of it. A proper test would take a very long time, since it would have to run the tubes until they died, switching on and off every hour or so. But with old-stock and the supplied Sovteks, I already like this slow start, it's almost spooky how smoothly the music blooms after 10-20 seconds, out of a totally silent background. For my own personal Paramounts, I'll probably convert them to 12AU7 or 6SN7 drivers to reduce the gain once we are done experimenting on them.
