Well, in an attempt to prove or disprove the advantages of using the emitter follower, I did some experiments this morning. I built the meter amp as shown in PRR's modified schematic above, with a couple of minor changes so I could use the parts I had on hand. The bias resistors were 1 meg instead of 2.2 megs. The filter cap was 22µF, and the other filter cap at the phantom divider was 33µF. I replaced the 3.6K trim pot with a 22K resistor so I could inject +20 dBu of audio (1kHz sine wave) without blowing up my meter. Oh yeah... I didn't have any TL072 opamps on hand so I used an LF353 which is pretty similar.
The power supply was a simple thrown-together rectifier feeding a small-ish pi filter (two 22µ and a 1.8K resistor), producing +48 V with about 50 mV of ripple.
In both cases audio was present at the junction of the four 1 meg resistors and the cap, but was about 63 dB below at about 5 mV RMS. More importantly, there was no measurable audio reaching the 30V rail in either case.
Oddly, on my scope I could barely detect a very slight sign of audio riding on top of my not-very-well-filtered 48V when using the emitter follower, which I did not see when using only the PRR divider. So could the divider with that big cap actually be better than the follower?
In any case, my normal phantom supply is well filtered and regulated, so the regulator action should help attenuate any audio that reaches the output, although it all seems so unlikely that the point is moot.
Thanks PRR for the tip.