You've jumped from GUEs averaging the bottom depth (you are diving over a bottom with bumps between (say) 80 and 90 ft, so you call it 85 ft*) to depth averaging the entire dive, for example 90 ft is the max depth of the dive, and you do a kind of linear multi-level dive, which you now call 45 ft). THAT is quite wrong, will give the wrong assumed N2 uptake for the dive, and has been shown time and again that it is incorrect. The reason it is wrong is because depth averaging is a linear process, but the on-gassing of N2 is not. You on-gas more at depth than you would think from just simple extrapolations from shallower. You can convince yourself of this quickly with any table: look at the NDL for some depth; now look at the NDL for twice that depth; do you get half the NDL? No, not by a long shot.
I can't give you an example of a "multilevel dive that would be "allowed" by depth averaging the tables but that would overstep the limits of (i.e. lead to deco in) Buhlmann with a GF hi of 85, for example?" because there is no way to validly "depth-average" the tables. It is just flailing in the dark.
*By the way, you'd have to call this 90 ft on tables anyway, so what's the point?
You're misrepresenting what GUE teaches about depth averaging when using tables. I'm going to assume you don't know, and that it's not just a straw man, so I'll explain.
First, your example is wrong: GUE has never suggested averaging the entire dive. To use your example: If you have a continuous ascent profile from 90ft to the surface (as if you're swimming along a sloping bottom), you're supposed to average the max depth and half the max depth (which is where GUE min deco ascent would start on a square profile). In this case the NDL limit on 32% nitrox according to the GUE tables would be 60 minutes (depth average 67.5ft, using 70ft table), rather than 40 minutes (at 90ft).
The other profile GUE uses for depth averaging is a multilevel bottom phase with a vertical ascent:
Let's say your visiting a wreck lying on a slope between 100ft and 70ft. In this case, you could plan to spend some time on the deep part of the wreck at 100ft, and then spend the rest of the bottom time on the shallower part, or swim gradually from the deep part to the shallower part. You could then use the average depth of the bottom phase for using the tables. If you plan to spend equal time at the deep part and the shallow part, you would use 85ft as the average, so it would give you 40 minutes (90ft table) instead of 30 minutes.
As long as it's not a sawtooth profile, or a progressively deeper profile, this works quite well, and can be done on the fly if you can reset the average depth on your bottom timer, or estimate your average depth (obviously more risky and task loading).
I'm not trying to say that this is as flexible or efficient as using a computer for multilevel dives (which most GUE divers, including me, are doing). But as long as these profiles are safe and within the limits of a computer dive, it's misleading to say that it's flailing in the dark. This is why I'm genuinely interested to see if there are examples of dives planned in this way that would overstep the limits of a typical dive computer (Buhlmann), to learn from it.