That would likely depends of what you're aiming to do with your track & it's a question of workflow too.
If you aim at just saving CPU & you plan to tweak your MIDI information or maybe add plugin automation, then you would like to FREEZE to be able to modify those later on.
If you are sure to be done with the sound & the MIDI & plugin automation then you could bounce the track to audio and it will save some CPU as well of course, but the difference between the methods is either you're done with the midi track or not.
In this particular case, Kirk is not "exporting" the track to audio ( that would mean to render an audio file from that track outside the DAW from Live's export option, which you can do as well ) but here he's recording the midi track output to a new audio track. This is the common workaround in Live since you don't have a direct "Bounce to Audio" feature like in Logic ( i.e ), you can do it in Live but you need to FREEZE the track & then FLATTEN the track.
Recording or re-sampling the output to a new audio track can also be used in a creative way or as a workflow because you could add new audio effects on the target audio track and record in "pre" or "post" Fx mode to create a dry or wet version of your midi track for example. You can even imagine to send the Midi output to an Audio Track A ( with effects ) & then send the Audio Track A output to an Audio Track B to get your final processed sound, like you would do when sending audio to an external device and getting it back into your mixing console in the analog world.
Live is very flexible in I/O settings to achieve very creative & cascading routing that can be very useful.
Hope that helps !