Today's first discovery: it appears prefixes are not the only things that use the run length limits...
I noticed that in row 95 of the Google Docs spreadsheet, we have "Graeh" that is a little way further along followed by "Groec". Seemed a bit weird since it's the same prefix, and yet "h" is nowhere near the end of the suffix list.
I took a look in the galaxy map, and "Graeh" is immediately followed by "Groeb" - i.e. it's jumped to the next infix.
On a hunch, I checked how long the run is for the "Ae" prefix, since I didn't think it was full-length. The answer? 12. How far into the suffix list is "h"? 12th position. Looks like infixes also use those run counts!
There aren't many infixes that also appear in the prefix list with non-full-length run counts, so it was quite a lucky spot really!
This does also explain how 35-length prefixes get "out of sync" (runs starting with suffixes halfway through the list) - they've previously gone through the non-full-length infixes at least once.
Still no closer to working out the discrepancy between the full-Z-slice length and how long it "should" be though. Clearly additional inspiration required.
Edit: OK, so I have a new theory about the continuation behaviour...
I think the behaviour we see (the later phonemes happening to continue when the prefix changes) is basically due to chance.
I believe each prefix keeps its own "counter" for how far through the list it's got. Since the majority of the prefixes are 35 long, this counter is in exactly the same place for all of them - thus, it appears to repeat.
That's also why it goes "odd" for non-full-length prefixes - its counter is further back because it hasn't had as many sectors, so it produces different names.