Wow Jakob that is a great article. Would you by any chance have the rest in the series to post?
Re paralleling to reduce noise: Sure it works, IF you assume you can ignore input current noise. The insight is that the signal goes up as N and the noise (since it is uncorrelated for each paralleled amp) goes up as root N, for a net root N improvement in S/N. This works for amplifiers in general. However, most of the time you get the same result with fewer parts by using larger geometry devices or paralleling input devices at least. Again, we assume that the source impedance is small, since o/w the current noise will eventually dominate.
Distortion, if it arises from a consistent mechanism (like "dead" [as PRR puts it] emitter resistance, hence a departure from good logging conformance) will clearly not be reduced by paralleling. However, if its effects are even-order distortion that at least can be reduced vastly by balanced topologies.
13700/NE5517 parts (the latter another casualty of the Philips Caen fire) are not very good but can have vastly improved distortion performance (even as singles) coaxed out of them by using the linearizing diodes and real differential drive (the app circuit in everybody's datasheet is a crude approximation to differential drive). Differential current drive is best but usually with just resistors running into the linearizing diodes, you get close to as good as it gets. Regarding noise, the sad thing, for these parts, is that a lot of the noise arises from the current mirrors and there is nothing to be done outside the chip to alleviate this. If they had provided pins coming out for the emitters of the Wilson cells you could add external resistors (which, in a current mirror reduce the effects of transistor voltage noise; unfortunately they reduce bandwidth too, and the lateral PNP's are already challenged itr). Of course that would be a bunch of pins and this is a cheap part.
I did a mid-fi product where I had to cheaply vary the gain from the satellite speaker with a single wire carrying a d.c. control voltage (commoned with the signal return for the speaker, so three wires in all in the cable). It was a four channel + subwoofer system and I used 5517s with a control current scheme that varied both the Iabc current and the linearizing diode current. It was complex but fairly cheap and worked way better for noise pickup and especially reliability, than running a million wires up and back from the satellite. S/N and distortion were not studio quality but not likely to generate user complaints. The product fairly bombed but not because of the gain control at least ;-).
Brad