Hi Angus and Geoff,
An interesting problem which I've just started to take a look at.
The first thing was to load it into a program other than Sekaiju, which is a great program, but it doesn't show exactly what's going on.
My preference is an ancient sequencer by Yamaha called XGworks (a sequencing program from the nineties).
This shows the uploaded file to consist of many track (or should that be channel) fragments in what can only be described as a seemingly random order.
See the image WC2-48-01.
If you observe the left of the image you will see that the channel assignment is in no order whatsoever (that I can discover).
(I can't get Sekaiju to do this. Can you?)
The first thing I observed was that there's no observable channel 1 usage in XGworks but there is in Sekaiju.
What I read into that is that all the meta events have been inserted into channel one instead of into the header track.
(XGworks has automatically reassigned all these and thus emptied channel one of data.)
These appear as shown in WC2-48-03.
Taking this and aligning all the channels together in the correct order has produced the image in WC2-48-02.
(You'll probably notice that XGworks has done some "tidying" of the lengths of some tracks where there are no MIDI events.)
This allows (me at least) to have a clearer overview of what's going on, and then to be able to look at a channel's worth of events more easily.
To be continued.