Here are my results.  I combined the 10 and the face cards in one row
since they're essentially equivalent.  And the data was generated by
doing 10,000 deals per upcard.

upcard         bust       17       18       19       20       21
natural
------      -------  -------  -------  -------  -------  -------
-------
     2   |   35.95%   13.32%   13.17%   13.34%   12.44%   11.78%
0.00%
     3   |   37.85%   13.41%   12.93%   11.69%   12.51%   11.61%
0.00%
     4   |   40.14%   13.24%   11.86%   12.03%   11.48%   11.25%
0.00%
     5   |   41.80%   12.82%   12.32%   11.41%   11.34%   10.31%
0.00%
     6   |   41.76%   17.65%   10.59%   10.36%   10.04%    9.60%
0.00%
     7   |   26.77%   36.23%   13.53%    8.29%    7.88%    7.30%
0.00%
     8   |   24.96%   12.84%   35.69%   12.73%    7.04%    6.74%
0.00%
     9   |   22.98%   12.27%   10.69%   35.47%   12.22%    6.37%
0.00%
    10   |   21.61%   11.21%   11.68%   11.23%   32.97%    3.55%
7.75%
   ace   |   11.60%   12.33%   13.03%   13.06%   13.43%    5.27%
31.28%

I do wonder, though, whether the deck is biased somehow when the
dealer deals him/herself since all the players are dealt beforehand.
For example, it seems a player is more likely to consume a sequence of
low cards through consecutive hits than a sequence of high cards.  So
depending on the number of players at the table, the cards the dealer
is likely to get may change somewhat.  And perhaps the dealer's odds
change further the deeper into the shoe the table gets.  Does anyone
know?

Eric

====

Are you interested in on-site Ruby training that uses well-designed,
real-world, hands-on exercises? http://LearnRuby.com