Which is why I did write "[...] if you ignore problems with noise an linearity" and mentioned problems with signals racing past the comparator trig values.

But from a theoretical point, the solution doesn't have a limit. The thing is that the concept it itself doesn't have a hard limit on number of supported inputs - it is just a question of how good you can control the quality of the input signals, the OP-Amps and the comparator levels. And you will need to consider a filter time on the comparators to avoid short spikes during rise and fall of input bits.

The situation here is that with good controlled input signals to the OP-Amps, the solution would scale quite well.

About your zero levels - it just a quality factor, and must be put in relation to the potential your summing OP-Amp us using.

When it comes to noise, you have a number of things to consider. The average voltage in relation to the potential of the summing Op-amp. And that multiple noisy signals when added together will have lower noise than the sum of the absolutes of the signals. And that an analog circuit will be affected by filter times in a similar way that a digital solution will be affected by variances in delays through different gates, which means the number of input values it can process/second will be affected by your control of the variances of the input signals. Different physical reasons, but similar end results.