Following this success, and with the intention of building a lift hill, I put the theory to work. If a derailed cart landing on a hill lands at the top of that hill, then why can't you derail off a ramp onto a higher one directly? I tested this, and results were extremely encouraging.
______
/
/
____/
Let the above be the horizontal view of a lift ramp. Each uphill is to the east (or west, or north, or south, depending on your configuration) and the tracks are EW tracks on the flat segments, and E-facing only, on each ramp.
A dwarf pushed the cart, and it rolled up all three hills, with a speed of 5 ticks per tile at the top, and 5 ticks spent on each observed ramp. This is strongly encouraging, and implies that this configuration very nearly turns a hill into a flat track, which would mean a dwarf could potentially push a cart up 100+ z-levels without additional same-level accelerators, or, realistically, one or two, given curves (and the desire for speed).
The most convenient thing about these configurations is they can push a cart to top speed and then send it to the top at that speed, making transit times to the top about the time it takes an unencumbered, average peasant to walk 4 tiles. I have not tested downwards pushes on the same route, but, of course, dedicated one-way tracks are simpler to manage for unguided routes, so a simple downwards hill would likely be better.
In summary, ELEVATOR!