I always assumed since every node holds its own mempool there is no guarentee the total 300mb for each node will be the same and this overflow of 300mb is a result of that. But don’t know for sure and not fact checked this is just the explanation that come to my mind when looking to that.