Using the SGF tree feature on a large database of more than 1.6 million games, I produced a large corner dictionary. The search region was an 11x11 square in the lower right corner of the board. Only moves which were played in at least 200 games were considered, in each position only the 30 most frequent continuations were considered (of course, usually there are much less), and at most 35 moves were considered.
After unpacking, you obtain a 14.3 MB SGF file.
The games in the database are around 60,000 pro games, and around 1.6 million games from 4d+ players on KGS. In particular, the quality of the moves will be mixed. The file is not a joseki dictionary, but rather an overview of moves played in practice.
Some information about the frequency and date period of the continuations is given.
The sgf file has 105406 moves (positions which can be reached via move sequences differing in the order of moves, and continuations from those nodes, occur multiple times, as determined by the SGF format). In comparison, Kogo's Joseki Dictionary has 62749 moves.
The RAM requirements for the computation were not exceedingly high (around 6 GB for Kombilo), so should be doable on a machine with 8 GB RAM or more. However, the pattern take quite some time (more than 24 hours on the 40 core machine which I used).