A pair of 6A3/2A3/6B4 will be sweating hard trying to do 7.5 watts, imo.
Best to double check specs to be sure.
Otoh, driven by a transformer on their grids, you can run A2 and push more on
peaks.
As far as going "balanced", the easiest method is to simply transformer couple each
stage! You already have the LS-10X for the input (overkill!), so you'd need something
between the input tube and the 6A3s. You could make a bunch of gain using a
cascode at the input and maybe only need a transformer to drive the 6A3 grids...
Of course, you'd probably want tubes on each half (positive and negative going leg) of
each transformer stage.
There's two different ways to run the driver to the 833s, one is to use a follower (unity gain)
like Gates does or to use a tube that makes gain. The difference comes in both the
control the driver has over the grids of the 833s and gain needed. The unity gain driver
stage has the advantage (usually) of "caring"
less about the transition from drawing
no current, to drawing a LOT of current when the grid of the 833 draws current.
A bit depends on the way the modulator tube is biased... is it Class 1 or Class 2? Some
tubes are "zero bias" so the grids are going to draw current from the git-go. Tubes that
are negative biased will draw nil current until one hits the "zero" and goes positive. So
that looks like a big impedance change. The driver goes from loafing along to grunt. The
cathode follower arrangement handles that well. But you need all of your voltage gain in advance
in the case of the follower
There are two main methods of doing balanced
circuitry. One is transformer coupled,
and then two signal paths in parallel... or using differential pairs (long tail pairs), or a combination
of these.
Oh, it is possible to make all of your (voltage) "gain" in the transformers - in that case the tubes then provide
the current gain to drive the next stage...
You can go balanced without any transformers using the diff pair method.
However making the diff pair
actually be balanced (IF that matters) requires more
than trimming resistors on the plates. Also the diff pairs may be sensitive to gain differences
between pairs of tubes or sections of tubes.
Yes, one could transformer couple diff pairs...
Oh IF there is an advantage to going transformer coupled it is that transformers (to the limit of
their internal balance) automaticially enforce "balance" between positive going and negative going
sides, whereas an entirely active signal path needs great care in the design to maintain true balance
and not have an asymmetry. (Of course for ham radio, a bit of less than fractional % perfection is
not important at all - in studio gear, it may be)
So, the main design considerations will be to work back from the output side (drive to the 833 grids)
to the input, and figure out how much gain (dB ---> volts of swing) is going to be needed. Then
fill in that black box... first step is how will you couple to the grids of the 833.
Anyhow the older mags and handbooks will have examples of entirely transformer coupled
speech amps...