OK, found the issue.
The input stage uses voltage-to-current NFB, which reduces considerably its input impedance, to about 10kohm.
In order to get a workable gain, the input trim was at about mid-rotation, which results in 34dB attenuation of the input signal before reaching the input stage. Of course, with such a small signal, noise becomes seriously perceptible.
I don't know what the designer's intent was, but with the input trim at its minimum resistance (minimum attenuation), the signal/noise ratio is good, but the global input impedance is about 20k (half of it due to a physical 10k resistor), which makes it unusable with a passive instrument.
Since I'm gonna use the Acetylene after a bunch of active pedals, it should work OK, provided I can tame the gain.