Also I found a nice paper of academic interest about sequence searching (not exactly your problem, but I couldn't resist as it has nice hashing of multidimensional tables at around section 4.3) It involves sliding a window over the target and storing the slopes of segments in very big hashes (if I understand as much as I read). I'd like to mess more with this but it's 3am here..

If you do not need an exhaustive list of all pattern s but just the most interesting ones, or statistically significant ones allowing for a number of sequence errors, biological code is more interesting for you. You could look for example at how they do RepeatFinder at TIGR if curious.

Also you could google about "hidden Markov", "interpolated Markov" or Viterbi which are used often to find hidden sequences or attempt predictions of what will come next in a sequence.