At a 7 tx/s creation rate, I think there should be no problem with mempools staying in sync prior to this PR, so your results make sense to me, for what it’s worth.
Do you know what’s causing the simulation to be unstable? Too much CPU signing/verifying txs? Or do the wallets get too large/slow, so tracking which utxos to spend is slow? Or is performance okay, and you just start getting inconsistent results between runs?