Thanks for answering.  Unfortunately, the regtest network is currently essentially unusable in the intended way once you've generated enough blocks to reach the first retarget interval.

I suppose this is a bigger deal for us in btcd because we have tests that exercise these types of things and do a lot of simulation testing using simnet (which is effectively like the regtest network without some of the things specific to the block tester tool and some additional logic to prevent address propagation and discovery).  That is a big reason we discovered this issue since we were comparing behavior during some simulations and noticed that Bitcoin Core was erroneously rejecting blocks due to having too low difficulty, when in fact the difficulty was correct.

I have to admit I'm disappointed that the answer is basically that Bitcoin Core doesn't feel like regression and simulation testing is important enough to warrant proper retarget behavior for it, but I appreciate the response nonetheless.