If the windings are on the same core, or if the secondaries were wired in parallel, it's split cathode feedback, most famously used by McIntosh but many others have used it to varying degrees. A common example these days being, put the cathode on the 8 or 16Ω tap, typically getting 5 or 10% voltage feedback, thus limiting the output stage's gain to about as much (below 20 or 10), and whatever the difference is, that's the local feedback ratio and therefore the reduction in distortion and output impedance.
As shown, the plate is a high impedance, so the bottom transformer generates all the voltage while the top transformer drops it all in reverse, doing nothing. Give or take what the two other grids are doing.
Tubes with suppressor grids, aren't worth any gain through them unless specifically made to. There are special multi-control types, 6HZ6 for example, which have a little gain this way. They have low perveance: IIRC, 6HZ6 has as large a cathode as 6V6's, but has a tiny fraction of its power output capability! Even if you had a power tube with significant g3 gain, it's shown as positive feedback, which is liable to make an oscillator. Small amounts aren't automatically out, but it's unlikely to be helpful.
Screen feedback gives a fractional triode mode ("ultralinear" at the classic 43% point; other tappings are possible), which reduces plate resistance, so will force some division between the transformers. The reduced transient screen voltage does reduce power output some (IIRC, the 43% figure is supposed to be a compromise between distortion and power capacity?), and large tappings (up to 100% triode mode) reduce the power output all the way to, well, that of a triode.
Tim