Don't use alignments here. Those are not optimized for finding exact matches at the rate you need.

The simplest method, in my opinion is to break the sequences into 11bp pieces, one piece per line, the sort and uniq rank them like so:

pieces.txt | sort | uniq -c | sort | head

As an example:

import sys
from Bio import SeqIO

# KMER size
SIZE = 11

recs = SeqIO.parse(sys.stdin, format="fasta")
for rec in recs:
    for i in range(0, len(rec.seq)-SIZE):
        seq = rec.seq[i:i+SIZE]
        print (seq)

then you can:

# Get some protein data
wget -nc http://ftp.ensembl.org/pub/current_fasta/accipiter_nisus/pep/Accipiter_nisus.Accipiter_nisus_ver1.0.pep.all.fa.gz

# Chop it into pieces
gunzip -c Accipiter_nisus.Accipiter_nisus_ver1.0.pep.all.fa.gz  | python chop.py > pieces.txt

# How many times did each piece occur
cat pieces.txt | sort | uniq -c | sort -rn > count.txt

looking at the end of the file:

tail count.txt

it prints:

  1 AAAAAAAAAGH
  1 AAAAAAAAAGD
  1 AAAAAAAAAFH
  1 AAAAAAAAAEK
  1 AAAAAAAAADL
  1 AAAAAAAAACR
  1 AAAAAAAAAAX
  1 AAAAAAAAAAN
  1 AAAAAAAAAAM
  1 AAAAAAAAAAD

the bottleneck is sorting, depending on the organism's size.

instead of sorting it as one file, split the counts.txt file into multiple files, sort each file in parallel, the merge sort the result