A pillar page should link to every genuine member of its cluster, and to none outside it, which means the “every versus closest” binary is the wrong frame. A pillar exists to be the cluster’s map, the page that organizes a topic and points to each piece that covers a part of it. For a true cluster, its genuine members are most or all of its spokes, so linking to “every” member and linking to “only the closest” turn out to be the same set, because every real spoke is a close, on-topic member. The pivot is real topical membership: link to what genuinely belongs to the cluster, and stop there. The choice dissolves once you see what the pillar is for.
The framing the binary misses is that a pillar is not a page rationing link equity to a chosen few, it is a hub mapping a topic. The advice to “link to fewer pages to concentrate equity” is meant for other situations, and misapplying it to a pillar defeats the pillar’s whole purpose, which is to connect the searcher and the crawler to the full set of pages that make up the topic. A pillar that links to only a handful of its spokes is an incomplete map, leaving real cluster members disconnected from their hub. The opposite error is just as wrong: padding the pillar with links to unrelated pages forced in to inflate the link count, which dilutes the cluster’s signal and points users somewhere they did not come for.
So the question to ask of each potential link is not “is this the closest” or “how many should I include,” it is “is this page a genuine member of this cluster.” If it covers a part of the pillar’s topic, it belongs and gets linked. If it does not, it stays off the pillar no matter how convenient an extra link would be.
For your next pillar, list the spokes that genuinely belong to the topic, link the pillar to all of them, and resist adding anything outside that set for the sake of link count. Link every genuine member, none outside, and the every-versus-closest dilemma goes away.