We often recommend to split large date ranges that use now into a large part that is cacheable and a smaller one that is not cacheable yet fast since it matches fewer values. Maybe we should look into automating this?