Logged In: YES 
user_id=80475

A few thoughts:

* If bisect provided an optional lt argument, it would still
not be thread-safe.  The code for the lt function can call
arbitrary Python code that runs through the eval-loop and is
hence subject to a thread-switch.

* An advantage of the class wrapper approach is that  the
prefix function gets computed just once per word.  This is
not a big deal for the specific case of [:3], but it could
be a significant benefit for other functions that are
expensive to compute (because repeated calls to bisect will
access the lt function more than once).

* My own approach to the particular use case would be to map
prefixes to a list of revwords or (revword, flag) pairs:
  d = dict(abb=['abbc'], abc=['abcaa', 'abcdd', 'abcss'],
abd=['abdf'])
This gives O(1) lookup time and limits the calls to the
prefix function to once per word.

Will leave this request open as it may yet be a good idea. 
My concern is that it will clutter the code and the API for
only a small benefit. 

Also, I'm looking at a more general purpose solution that
would make this feature request moot.  This idea is to
create a fast comparison decorator class used like this:

   dwordlist = map(ComparisonDecorator(lambda x: x[:3]),
wordlist)
   lp = bisect_left(dwordlist, given_rev_word)
   rp = bisect_right(dwordlist, given_rev_word)