If you want to avoid heap fragmentation (especially with server-side code), and thus avoid allocating huge byte arrays, you can do streaming, something like this:

        using (SqlConnection cnx = new SqlConnection(@"your connection string"))
        {
            cnx.Open();
            using (SqlCommand cmd = cnx.CreateCommand())
            {
                cmd.CommandText = "SELECT [File] FROM [com].[catalog1] WHERE [FileName] = @filename";
                cmd.Parameters.AddWithValue("filename", Request.QueryString["filename"]);

                // sequential access is used for raw access
                using (SqlDataReader reader = cmd.ExecuteReader(CommandBehavior.SequentialAccess))
                {
                    if (reader.Read())
                    {
                        // must be lower than 85K (to avoid Large Object Heap fragmentation)
                        byte[] chunk = new byte[0x10000]; 
                        long read;
                        long offset = 0;
                        do
                        {
                            read = reader.GetBytes(0, offset, chunk, 0, chunk.Length);
                            if (read > 0)
                            {
                                Response.OutputStream.Write(chunk, 0, (int)read);
                                offset += read;
                            }
                        }
                        while (read > 0);
                        Response.AddHeader("Content-Type", "application/pdf");
                        Response.AddHeader("Content-Disposition", "inline; filename=" + Request.QueryString["filename"] + ".pdf");                        
                    }
                }
            }
        }

And if you serve the same file often, then it's better to write this file directly to some Cache directory on the server and then re-use it for subsequent request, with the Response.TransmitFile API which is the best in terms of performance (uses kernel mode if possible).