There are plenty of things going on in a real neocortex that are not essential to intelligence and Numenta was working hard from the start to pick out the important ones at an appropriate level of modeling. This distinction becomes especially important in contrast to things like Henry Markram's projects, where people seem to be happily spending major-multi-million-euro sums of public funds to run poorly understood simulations with insufficient detail on a really big computation cluster.
RNN:s are crude compared to the HTM as well as to the more recent CLA model, but they do share for instance the formulation as a dynamical system with recurrence. Some of publications coming from people at Numenta have made direct comparison between the CLA to for instance LSTM-based networks, and in my opinion they are entirely reasonable to do that. They are all suited to sequence modelling.
Having said that, I agree the comparison to RNN:s isn't to be made without understanding some of what goes into the HTM or CLA, and I agree that Hawkins emphasizes biological intelligence.