I've put up a python bloom filter implementation at http://stromberg.dnsalias.org/~strombrg/drs-bloom-filter/
It's in pure python, has good hash functions, good automated tests, a selection of backends (disk, array, mmap, more) and more intuitive arguments to the __init__ method, so you can specify an ideal number of elements and desired maximum error rate, instead of somewhat ethereal, datastructure-specific tunables.
