From the user perspective, they might be interested in tuning the plan cache properly to accommodate the number of entity queries that require to be compiled. If we break it into multiple steps, it might be difficult to focus or figure out which one to observe.

It's true that for some queries, the filters could influence the plan, but if the statistics tell the user that this particular HQL query takes too long to compile, either if using filters or not, I think the user will try to address it at the plan cache level anyway, rather than on a per-query basis.