We only have existing experiments for CPLEX but I'd also test if this helps with SoPlex. I didn't find an option quickly but assume there is one. We could have a look at the OSI interface to figure this out.

Sounds good. Is this for CPLEX only (like the thesis, if I recall correctly), or are there also similar settings for SoPlex that should be considered?

The master thesis by Renato Farruggio found that presolving the LPs of potential heuristics hurts more than it helps: https://ai.dmi.unibas.ch/papers/theses/farruggio-master-24.pdf

We should switch it off for heuristics like this.

We could also look for other improvements in the thesis results and either handle them here or create new issues for them.