BLOG_DOCS
----------
doc_id number PRIMARY KEY
doc_title varchar2(25) NOT NULL
doc_author varchar2(25) NOT NULL
doc_data CLOB NOT NULL

create index blog_doc_txt_idx_ft on blog_docs(doc_data)
  indextype is ctxsys.context
  parameters (
  'WORDLIST SubStrWordListPref
   STORAGE DocsStorage
   DATASTORE MultiColumnDataStore
   sync (on commit)'
);

SELECT *
FROM
  (SELECT tmp.*,
          rownum rn
   FROM
     (SELECT /*+ FIRST_ROWS */  *
      FROM blog_docs
      WHERE contains(doc_data, 'richard', 1) > 0
      ORDER BY score(1) DESC) tmp
   WHERE rownum <= 20 )
WHERE rn > 10;

You should use the FETCH FIRST syntax for top-N queries since 12c:

SELECT *
FROM blog_docs
WHERE contains(doc_data, 'richard', 1) > 0
ORDER BY score(1) DESC
OFFSET 10 ROWS
FETCH FIRST 10 ROWS ONLY;

It's also better to use the "seek" method rows to get the rows for pages 2 onwards. This is where you save the value(s) from the last row on page N - 1 to filter the data.

Which would look something like:

SELECT d.*, score(1)
FROM blog_docs d
WHERE contains(doc_data, 'richard', 1) > 0
AND   score(1) <= :last_score
AND   doc_id > :last_id
ORDER BY score(1) DESC, doc_id
FETCH FIRST 10 ROWS ONLY;

This helps because the query only processes the N rows you want. With OFFSET it'll also read the M before this.

See more about this at https://asktom.oracle.com/pls/apex/asktom.search?tag=comparing-pagination-methods-offset-vs-seek

When it comes to analyzing your queries, the DBAs should really be checking row estimates are accurate and how much work (buffers/consisten gets/logical IO) the query actually does. Focusing on (CPU) cost alone is kinda meaningless.

https://asktom.oracle.com/pls/apex/asktom.search?tag=explain-plan-cardinality-and-cost