Is it possible to collect all redundant fan-garbage nodes on the abstract Lamping algorithm?

The Lemping Abstract Algorithm is an efficient way to reduce a member class in pure lambda calculus. I noticed, as asked in the previous question , that only the abstract algorithm leaves piles of redundant nodes after pruning, which is responsible for the exponential slowdown when shrinking some terms. Marco Solieri responded by pointing out that this is common and suggested a method to fix this problem by converting the "safe" fan assemblies into flat wires - ie. Garbage collection.

This solution is not perfect, though, as it still leaves a lot of redundant nodes in the final graph. My question is, is it possible to collect all the redundant fan-garbage nodes when performing the abstract Lamping algorithm ?