SQL query with ROW_NUMBER is ineffective in LINQ

I have this query in SQL:

SELECT * FROM (
    SELECT ROW_NUMBER() OVER (PARTITION BY p.category ORDER BY (
        SELECT AVG(CAST(r.users AS NUMERIC)) FROM description r WHERE r.company = p.id
    ) DESC)  AS rownb, p.* FROM company p 
) rs WHERE rownb <= 2

      

        
        
        
      

    

The farthest I came up with converting this query to LINQ:

Companies
    .SelectMany(r => 
        Descriptions
            .GroupBy(t => t.Company)
            .Select(t => new { 
                Average = t.Average(a => (double)a.Users), 
                Company = t.Key })
            .OrderByDescending(t => t.Average)
        , (p, r) => new { Companies = p, Descriptions = r })
    .Where(t => t.Companies.Id == t.Descriptions.Company)
    .GroupBy(t => t.Companies.Category)
    .Select(t => t.Take(2))
    .SelectMany(t => t)
    .Select(t => new { t.Companies.Name, t.Descriptions.Average, t.Companies.Category})
    .OrderBy(t => t.Category)

      

        
        
        
      

    

But the problem is performance. Although the cost of the SQL query is 28% (relative to the package), the LINQ query is 72%.

I've already replaced Join in SelectMany in LINQ, which reduced the cost by 20%. But now I don't know how to optimize this query.

Also, my understanding is that LINQ is missing ROW_NUMBER.

I am using LINQPad to validate the resulting SQL query.

Question: Is ROW_NUMBER responsible for this performance difference? Is it possible to optimize a LINQ query?