Apple's approach goes a fair bit beyond simple servo control (which only tends to work well for truly low-frequency content.
While Apple hasn't disclosed much about the particular details, it's bound to be in the same vein as Klippel's work, which doesn't use feedback as much as it monitors a few states as input to a feed-forward error-correcting mechanism. The net result is that distortion is lower, and it's much less limited with regards to the frequency range you can support.
[1] http://www.audioxpress.com/article/louder-faithful-and-consi...
[2] https://www.klippel.de/fileadmin/_migrated/content_uploads/A...
They did not have wifi or an always-on microphone, true, but the former is easily remedied (use a Raspberry Pi Zero or something similar) while the latter is more of an advantage than a disadvantage.
[1] http://www.extra.research.philips.com/hera/people/aarts/_Phi...
[2] http://hifipig.com/philips-motional-feedback-speakers/