We just spent a few months attempting such a feat. We sought very low noise tubes, did a lot of testing of various resistors to see if they contributed to noise, ran a few different high voltage shunt regulator options, paralleled tubes to reduce the 1/F noise, and tried a whole bunch of other stuff. The end result used 4 6C45pis per channel. It sounded amazing, but we just couldn't get an acceptable noise floor.