I'm late to the thread, but I used to be a test engineer (decades ago) at Analog Devices and had their analog multipliers as one of my product lines. (essentially a Gilbert cell in a configuration that allowed the product of two +/- 10 V signals to be produced. Termed a "four quadrant" multiplier.) The Gilbert cell was a great innovation and is really best implemented in a single IC where, as Amir said, matching is inherently quite close.Countless is highly suspect FWIW. Yes after 40+ years you might not hear some of the issues that show up in measurements, but this one isn't good at all. Imagine pairing this with extremely high sensitivity speakers, I'm guessing you would notice then when compared to another more competent unit . If you're happy that's all that matters, right?
So, I haven't looked at a AD534 data sheet in a while, but there is no way they are going to support linearities that will result in >100 db SINAD. Their main use was in analog control and computation circuits of various types.
Exceeding that device's specs with discrete transistors is a fool's mission.
I have hacked with them and produced a simple modulator/demodulator similar to a super heterodyne receiver (in the audio and near ultrasonic band) and it was a fun demonstration but essentially useless.
Barry Gilbert was/is a down to earth and very smart guy.