An alternative to filtering the top 10 hits in a script after running BLAST is to ask BLAST to do this for you (using argument -maxtargetseqs). This way the output file will be smaller and the BLAST job should be faster to run.

Also, if you don't like XML and really just want the sequence, specify tabular output with just the matched sequence (field name sseq).

This is a good guide in doing what you want.

You may want to parse like this for getting the sequence.

    from Bio.Blast import NCBIXM

    blast_records = NCBIXML.parse(blast_handle)

    save_file = open("/path/to/file/my_fasta_seq.fasta", 'w')

    for i,blast_record in enumerate(blast_records):
      if i==10: break
      for alignment in blast_record.alignments:
          for hsp in alignment.hsps:
                save_file.write('>%s\n' % (hsp.query))
    save_file.close()

Update: fixed some errors