This looks as a more-or-less recent change in Boost regression scripts, since I've added "using python" to our user-config.jam some time ago (I don't remember exactly but it was not later than half of year ago), but tests started failing just about a week ago (well, maybe several weeks ago). Does this conclusion looks right for you?

If I'm right, we have to do something with that. This is not strictly Android-related issue, it is another cross-build testing problem of Boost.