A couple of things are be happening here:
As configured in the "Darling" circuit, the 12SL7 driver stage has a voltage gain of 43. The plate current is around 1.1 mA, and the voltage across the cathode resistor is 1.1 vdc. So, any input signal above 1.1 volts + or - peak will drive the stage into saturation, and distortion will result.
The 1626s have a plate current between 25 and 27 mA. Their 1K cathode resistors will have a DC voltage of 25 to 27 VDC across them, so any grid drive voltage greater than +/- 25 - 27 V peak, will drive them into saturation, and cause distortion.
Since the driver stage has a voltage gain of 43, then the maximum input signal to its grid, should be no greater than between 0.58 and 0.63 volts +/- peak, in order to limit its signal output to the 1626s' grids to 25-27 peak volts maximum, as anything greater than that will result in distortion.
So, Anything greater than 0.58 and 0.63 volts peak drive signal to the 12SL7 grids will cause the distortion to rapidly appear.
With only 3/4 watts rms maximul power output, even 93 dB speakers aren't going to rattle the windows and shake the floors
Maybe with some very efficient 103 db or greater, speakers, but then those tend to be large and very expensive
Just think of it in automotive terms: The tiny "Smart Car" isn't capable of doing zero to sixty in two seconds. Maybe, in two minutes, would be more likely. A Lamborgini can do 0 to 60 in 2 seconds, and perhaps even less time.
The 750 mW Darling is like the "Smart Car", and a hefty 100WPC amp is more like the Lambo.
Every +10 dB of power efficiency is effectively a tenfold increase (or a decrease for -10 dB) in percieved SPL (sound pressure level),
With typical small bookshelf speakers, with 83 dB efficiency, they would sound like 75 mW. into 93 dB efficient speakers Pretty weak....
With 103 dB efficient speakers, then they would sound like 7.5 watts into 93 dB speakers. Or, 75 watts into 83 dB speakers.
Your signal source probably has a line level output of +/- 2 volts which would drive the amp into distortion at about "halfway up" volume control settings.
One simple fix for this would be to connect a 200-220K resistor in series between the RCA input jack center to the "hot" end of the volume control, on each channel. THis would effectively reduce the maximum input gain by 10 dB, or 0.316 of the device's output signal.
I've used this -10 dB padding approach with a couple of overly "hot" line stages, and it worked very well.
Due to the tube's Miller Capacitance, this could very slightly affect the highest octave (10K-20KHz) frequency response, but it should not be readily apparent.
HTH
/ed B in NC