Right now the Classic engine has an optimization to early terminate the accumulator processing when the accumulator will no longer be affected, for example, $first and $firstN. This optimization is missing in SBE, which may cause unnecessary overhead including accumulator argument evaluation, extra VM calls and so on.