Some more testing and mods.
Measured efficiency of the PA only when RV1 adjusted for 10W output on each band (current measured through the PA-PWR line, current is zero during receive):
Changed T11 to BN43-3312 same wire and turns ratio, 2 primary to 4 secondary.
Power measured using same method as above with BN43-3312 no cap across the primary, each band set to 10W by adjusting RV1.
Much better and finals are cooler but still get hot with key down for 30 seconds. Some acceptable numbers there but why is efficiency increasing with higher frequency?
Note that power efficiency measurements are for the finals only.