The tubes are indeed GE, made in Owensboro Kentucky for the military (milspec). Indeed they are smaller but that's because the tube had become a GA (straight sided glass) rather than the older G (coke bottle). The Russians and Chinese never made that final step so their's are all G.
I had quite a time repairing/modifying and re-tubing a pair of early MA-1 amps for one of my customers a year ago. I bought enough tubes for both of his amps and some to spare. I had added cathode resistors (as per Atmasphere's recommendation) which allowed me to check the current sharing of the tubes. With the resistors in place this is very easy to do and quite alarming. I put in unmatched tubes from the same lot and found that the sharing of current from one tube to the other was just horrible. I was not surprised as this tube was not designed to be biased from a fixed source. It's most common use was as a pass tube in a regulated power supply where the grid bias is controlled by the error amp and varies over a wide range as needed. This is also true of the 6C33 (triple nipple tube) which was designed to do the same. Therefore, I can't rightly blame the tubes.
The result of this study was that for a given negative bias voltage the tube currents varied more than 2 to 1. This presents a problem. I've spoken to Ralph Karsten about this and he doesn't mind the imbalance in tubes as long as one bank is close enough to the other bank be within the range of the DC balance control. However, tubes in a bank might not be sharing the work equally. It's like money, if each bank of tubes adds up to $100 that could be 5 $20 bills (5 equal tubes) or a bunch of $1 bills making 5 stacks (tubes) that add up to 100 (e.g. 10, 20, 16, 14 and 40). That also adds up to $100. There's really no way of knowing unless you can determine the current of each tube individually. Further, the more off center you have to adjust the balance pot for 0 DC output, the more the banks are unequal as a whole. One bank may be biased at -60 and the other at -70 (or worse). That grid difference will raise even ordered distortion and cause the top or bottom of the wave to clip prematurely.The balance pot in the very early MA-1's I worked on had a very wide range.
Of course you can't determine any of this on a Hickok tube tester. I had to build a test fixture with very low drift regulated power supplies so that the measurements from one batch to another would be consistent. Plate voltage was held to better than 1% as was grid bias. I also tested them at the precise voltage and current that they would see in operation. There is no way to use double point (bias and gm) matching and get the accuracy required. BTW, I see lots of tube sellers claiming "double" and even "triple" point matching but they still aren't doing it right as several members have reported here. Ask them how double point matching works and why it is valid. I bet most of them don't know. Ask them if they measure Gm at a fixed grid bias or a fixed cathode current. So far all of them get that wrong too. They just jumped in on the research I did in 1982 when I started RAM Tubes.
As to George's cost concern; I can give guys with the big amps a break on matched set of 24. The small tubes don't require replacement nearly as often and don't cost much anyway. If you want to buy the small (driver) tubes at the same time let's call it $800 all together.
I have looked at pictures of several Atmasphere amps listed on Audiogon and could easily see the unequal tube wear. Just look at the tops of the tubes and note the size of the remaining gettering. When used at unequal currents for a long time the hotter tubes will have very little gettering left and the cool tubes may look like new.
Lastly, I have great respect for Ralph Karsten and his work. He has always been helpful with my questions about his amplifiers and he's a dedicated, no BS guy. That's rare in this industry.
