[Radio Wrangler]

Might be time for trying a bit of freezer spray?

Yes, transistors can have their RF gain and noise ruined, with seemingly lesser effect on DC conditions. It's one of the symptoms of something that has had vbe reverse biassed into avalanche, but not with enough current to completely kill it.

A decoupler/blocker gone higher resistance than normal in this circuit could spoil the gain.
So Q20 is running 36 mA and Q19 8.4mA accorging to the emitter voltages and resistor values. 1.8v drop on 51 Ohms says 35.29mA which fits.

The circuit looks a bit naive. There seem to be no anti-parasitic oscillation components which I'd expect if some lively devices were used with plenty of available gain at RF. However, this is a relatively early Japanese instrument, and the Japanese were evolving a design/manufacturing style of carefully optimising component choice to minimise component count.
Around that time I was working at HP and we had issues with finding some, er, excessively familiar circuitry in competing instruments from a Japanese firm, though not T-R and in an instrument a lot more specialised than spectrum analysers.... particularly in some areas which were the work of another forum member. I think at the time he said he felt 'sincerely flattered'. Did I say he has a dry sense of humour?

Anyway, it was a time when Japanese test gear manufacturers were just moving into more complex instruments that basic sig gens, and they were feeling their way around and learning.

If you swap Q19 or Q20, you may have to add some parasitic stoppers.

David