My experience gained from testing thousands of EL-34's, 6550's, KT-88's, and other tubes has produced a wide bell curve of the two most important parameters, Bias and Transconductance. The center of this curve is the "bogey" value that the manufacturer is trying to hit, and the ends show the cutoff points of what he is willing to allow out of the factory. The tubes at the ends are still perfectly functional, but may not work well in all amplifiers, and will certainly not work well if mated with another from the opposite end of the curve.
Some amplifiers do not have enough range on the bias pots to handle the range of tubes so that selection-must be made from a particular part of the curve to ever achieve bias.
In the "good old days," I'm told, the spread of values was much less, allowing reasonable performance with random selection. Today, random selection is strongly not recommended-I'll tell you why.
Looking at bias voltage in a typical batch of EL-34's from the best producer, it will typically range from -32 to -42 volts for 50mA of plate current. At a typical transconductance for 7000 umhos (7 mA/volt) the 10 volt difference will cause a 70 mA bias difference. Obviously, one tube will be on and the other cut off if they are biased with the same adjustment pot. I match the bias voltage to. 0.5 volts or better, and thus have only a 3.5 mA (0.3 x 7 mA/v) difference in plate currents, which is only 4% maximum at 50 mA.
Now, those of you with individual bias adjustments are probably thinking that you've got it all handled because you can adjust for these different grid voltages. Well, you can-but in doing so, you will create a greater problem.
Say you put a -32 volt tube in one socket of a push-pull pair and a -42v in the other and bias them so that their currents are equal. Unless you can also adjust the a.c. balance of the driver to provide 32v of drive to one and 42v to the other, you will have very unsymmetrical clipping and reduced power output.
In most amplifiers with a.c. balance, the range of adjustment is nowhere near this, as that adjustment is meant to balance the driver and not current for widely unmatched output tubes.
In our modern systems, larger amplifiers, multiple pairs are used in push-pull parallel to achieve higher powers. It takes four of the popular output tubes to safely attain 100 watts, and here matching is even more critical.
To fully appreciate the problem, you must know a bit about the operation of a tube output stage. In push-pull parallel, we have four (or more) tubes connected so that two (in parallel) handle the positive half of the output signal, and the other two handle the other half. Even if you do find two tubes which are 5v off, there will be 35 mA different in bias current, which is still intolerable.