3 factors make Kipchoge's marathon not comparable to his best 10000m time or ability.
1) rarely is a marathon run on a course with so few corners and turns. Dubai is one with very few corners and turns, but even it has 2 hairpin turns if I remember correctly. Never understood why Dubai didn't figure out a better way to eliminate those. London has a couple dozen corners and turns. Berlin has many too despite being flat. And all 3 of those courses are considered 3 of the fastest marathon courses in the world. So if Kipchoge runs a marathon on a course that is more comparable to the cornerless running on the track, his marathon should appear to be disproportionally faster than his best track times. When we compare Bekele on the track to Bekele's best marathon, Bekele never ran his best marathon on such a cornerless route as Kipchoge did yesterday. So it's not that Kipchoge underperformed on the track, but that the favorable course sans corners and turns allowed his marathon time to be disproportionally faster.
2) the pace car and phalanx of relay-style pacers created ideal drafting for Kipchoge. Even if that only contributed a 1 or 2 second difference per mile, that's still big when trying to run 7 seconds per mile faster than the WR.
3) more importantly, the pace car and laser green lines created even splits. I dare say (and I haven't checked this) that there has never been a sub-2:05 marathon ever run with more even splits for 5K or even 1K than what Kipchoge ran last night. Nike virtually eliminated even the most minor surging so that no energy was wasted. Except for the last 7km when Kipchoge strained to not fade, I imagine the first 35K had 1K and 5K splits that were more even than any other sub-2:05 marathon ever run. Someone could fact check this by finding the difference between Kipchoge's slowest and fastest splits last night (for the first 35K) and comparing it to the same of any sub-2:05 marathon ever run. I'd be curious to see it.