> How does the sound of the 6L7-based compressors compare to the Fairchild / Sta-level / triode based vari-mu stuff?
Probably depends MUCH more on the transformers and the time constants than the tubes.
> I am under the assumption that using the 6L7s will cause there to be less issues with control voltage leakage??
Don't see why.
> I am considering paralleling the 6L7s to reduce noise and get more output.
About 6dB increase in total dynamic range for each doubling of tubes. If you run a post-amp, like about everything except the Fairchild, then output of the vari-gain stage is typically not a big deal. And noise is not a big deal except for very wide dynamic range program input. So mostly the gain from paralleling is not worth the cost. If you don't use a post-amp, a la Fairchild, output level becomes THE design factor, and even a pair of 6L6 (yes, they could be used as vari-gain) will not give pro-like output level while doing deep gain reduction. The Fairchild uses a lot of pretty sturdy tubes, and still will not give the huge output levels demanded in broadcast limiters (which also drove long transmitter lines).
> current through a center tap. it will cancel out, because the current is flowing in opposite directions, but ONLY if the DCR of both sides of the center tap are the same.
For pentodes, DCR-match is not critical; for triodes it has little effect.
But you WILL have trouble with current balance. Say you pick or trim two tubes (or tube-halves) to equal idle current. Now start taking the grids negative to reduce gain. No two tubes (not even halves of the same bottle) match exactly over all grid voltages. Say idle at 10mA each, take the grid voltage down a bit, one tube will be say 5mA and the other 4.5mA (it could be worse than 10%). That's a 0.5mA unbalance. Not a problem for a 5K 25 watt output transformer, but could be bad news on a 20K 0.5 watt core.
> plate resistors and caps to keep away DC from the transformer
Nope. You can easily trim-out the static idle unbalance. You can't easily(*) trim or select away the unbalance shifts as you drive the grids negative and current from 10mA toward 0.1mA.
And there is NO way for the system to know the difference between a 0.5mA balance-shift and a 0.5mA signal transient. Since typical limiter time-constants are equivalent to the rise of a musical bass note (2mS attack is like the start of a 83Hz tone) we can't filter it away. We have to balance the tubes all the way down the current range, and accept that it can never be perfect.
Assuming the attack time can be as low as 5mS, and that you want frequency response flat to below 30Hz, while the coupling caps will block DC unbalance they will not block unbalance during limiter attack. You will have thump and you can have excessive unbalanced current which will eat-up your bass overload margin.
(*)Programming a complementary bias trim is difficult because of the high impedance of most tube limiter time networks. Also the currents could be 5mA:4.5Ma and then 1mA:1.2mA... not one tube consistently higher than the other. I suppose these days we could program a DSP to test the tubes after power-up and warm-up, memorize their curves, and calculate the right grid voltage for each tube for any desired GR. But if we go this far, we are well along the path to much more complicated limiting algorithms (a la CBS patent) and even distortion-cancellation. Pretty soon the tubes are just decorative heaters, the funky elegance of the tube limiter lost.
> I think a regulated power supply would be a good idea.
For a simple sidechain and broadcast-standard limiting level, you need to regulate, only the vari-gain stage (and only the G2 of pentodes). The output level rises until the gain is reduced enough. The amount of grid voltage needed for any given gain reduction is a function of plate/G2 voltage. 12AU7 working at 100V on the plates will give a lot of gain reduction at -7V on the grids; if run at 300V on the plates it needs -15 or -20V on the grids for similar gain reduction. Or simply: 1dB output level accuracy needs 10% supply regulation. In systems that assume fairly steady wall-power, or that don't need consistent limiting level, unregulated supplies worked fine. +/-20% line variation is +/-2dB output limiting level variation: if you just don't want speech to splatt, that's plenty good. If the wall wandered more than 20%, the tube heaters would be unhappy. If the limiter feeds a loudspeaker amp on the same wall power, the limiter "should" track wall-power wanderings, because the power-amp max output will be shifting too. Finally if you set the limiter by-ear every time you use it, exact limiting level consistency is pointless unless you also have large wall-power variations during a take or mix. (And if you do, that may be more a problem for other gear than for the limiter.)
If you try to run the vari-gain stage on a low voltage dropped from a high voltage through a resistor, you get a different type of limiter. As vari-gain stage gain drops, stage current drops, voltage drop in the resistor is decreased, supply voltage to the stage rises. This works against you, but is not necessarily bad. Many-many tube limiters work with high-value plate resistors and the plates soar from 60V at idle to 200V at -6dB GR to 230V in deep GR. This softens the knee but can be somewhat compensated by increased gain to the rectifier and increased rectifier threshold voltage.