To take the position that a single stage of 5687 can drive a SE, leaves a fairly simple( for all you newly created CCS experts ) long-tail-pair/differential amp phase splitter to run it from.

Since we're probably going to make our own custom OPT of between 5 and 7k a-a for loading the 2A3's it is no big deal to put in a pair of 10% E-Linear driver taps on the plate winding and again run with a super-simple single stage L-C B+.

and then ther's the op point. Lower voltage means you can run more current w/o exceeding the plate rating. The input voltage requirement will be lower, but so will the phase splitter's ability to generate it. A -50 volt bias indicates a drive requirement of 100V p-p, which is possible with a resistive loaded 5687. A DC-tolerant PP plate choke might be indicated here in order to achieve maximum output voltage swing on min B+.

With a PP choke, I'd be quite tempted to use a 6BX7 with mu=10. I am getting IT ideas as well. Phase splitt with an input tube working as a split-load. 12B4 will give performance like a 1:1+1 IT in that position. Feed a second pair of 12B4 riding a string of LED as a common cathode load bias arrangement. Load with the Lundahl 1660 PP:PP and you could even get some grid current with the 2A3 power stage. I bet it would overlaod very gently with such a driving arrangement.

Guinevere will comfortably supply 10-15 vrms to the imput...
regards,
Douglas