为什么窗口聚合函数的逻辑读取这么高?

I've found that in execution plans using common subexpression spools that the reported logical reads get quite high for large tables.I've found that in execution plans using common