With most audio power tubes, with most driver designs, you will have ample drive if the driver is fed a supply nearly as high as the screens.
G1 swing needed is approximately G2 voltage divided by Mu(g2) of the power tubes. Mu(g2) of EL34, 6L6, 6550, 6V6, KT88 etc is about 10 (EL84=18, 6146=4). A decent driver scheme can deliver 20% of supply as signal swing. Therefore EL34..KT88 will have more-than-enuff (EL84 tends to too-much, 6146 tends to fall short).
Now you ask if you want to run 12AT7 at 575V-100V= 475V. Not all of this will appear across the 12AT7, so we may not violate the 300V-330V rating. However "Silverface" to me means long-tail driver. We often drop 100V in the tail. That leaves 375V across tube and 47K resistor. The worst-case for tube dissipation is half-voltage, 187V across tube (OK!) and 187V across resistor. Resistor and tube dissipation are equal at 0.75V each. I think this is OK for 12AT7, clearly the resistor needs to be far larger than a default 1/2W part.
> 47k, 330k
Is 330K the power tube grid resistor? Note that MAX on 6550 and KT88 in fix-bias is 50K (100K for smaller tubes, often 500K in Self-bias). Yes, every g-amp maker violates this. However if there is a history of killed tubes, I would aim lower. With 47K plate load, 100K grid resistor will still drive well, and 100K will bleed-off a lot more grid-trouble leakage without excess un-biasing than 330K.
Agree that at this voltage, Ultra-Linear connection is worth consideration. When tube current is high, screen voltage drops. It is less likely to go nuts and squirt more current than is safe.