I'm not sure which mental model is correct, but if you think the value of the YC ecosystem scales at Θ(N²)...
The correct scaling for social networks is almost always n log(n). See http://www.dtc.umn.edu/~odlyzko/doc/metcalfe.pdf for a collection of arguments using various measures that all conclude that that is the right scaling laws.
Social Capital is not fixed, but Dunbar's number does provide an upper bound to it.
If social capital is mostly fixed, then this is a zero-sum situation and increasing batch sizes is bad for each company, even if it's better for YC as a whole.
Another mental model is that social capital is generated by YC companies themselves, as they e.g. network with one another, trade advice and other resources, and otherwise collectively learn how to build better startups.
I'm not sure which mental model is correct, but if you think the value of the YC ecosystem scales at Θ(N²), then a bigger batch can better for each YC company, even if they're get less 'direct' support from YC itself.