Well, high risk for sure! For a 3k load impedance, 3.5 watts is about +/-150v peak, i.e. 300v pk-pk.
But of course that means a B+ margin of 150v. Add in 275v plate to cathode, 55v bias, and 175v for the driver plate (direct coupled) and you need a 655v power supply. Add 10% margin for a shunt reg supply and it's 725v.
Now you might get 470v from the Paramount supply, so (scaling everything down) maybe 1.5 watts from a 2A3 would be possible. It doesn't sound like a big-selling product to me :^) but maybe an interesting experiment.
What would be that magical 1.5 watt op with a 3k opt, sr, dc and c4sed? I could go lower power if required, maybe as low as half a watt but 1.5 would be awesome.
Also, I can definitely skip the driver section so there's no need to calculate voltages for direct coupling nor currents for a driver. Just a simple SR, C4sed 2A3 spud for 3k opt and around a thousand milliwatts.