If you want 3 wide, I'd go with rollers. With some design effort you should be able to get more than one minecart in a loop falling on each tile and get it down to something like 11-12 ticks per strike, which is the time it takes for a minecart to begin falling after striking the wall opposite an open shaft.
Does anyone know if a column of falling minecarts still compact into a single tile, one above where they land? The last time I messed with dropping minecarts in shafts, they would all accumulate 1 tile above the landing, (pretty sure they were staying as projectiles), and materialize one at a time at the bottom, with the next one arriving only after the previous one was cleared.
If that still happens AND they retain their falling inertia, (which would be quite the combination of bugs), then you could achieve a faster strike rate by leaving the virtual carts stacked above the strike location.
Ooh, here's another idea: drop the minecarts from multiple drop stations at different z-levels above the target tile! It would be even more work to get the timing right, but it only takes what, 5-6 ticks? for a highest speed roller to clear the landing tile. If you had multiple drop stations above each x/y position, the the time to clear the drop location would be your rate limiting factor.
For this bug, it sounds like minecart mass matters. Eurgh. Well, then, since you're falling and not doing horizontal collisions, load those gold minecarts full of gold minecarts full of gold minecarts... May as well turn it into a cloud-of-body-parts generator, since you're already going this far to abuse visitors.