A very limited-use variant speed reduction method - the difference brake:
#### ####
#╔╗# #╔╗#
in=╬╝# =╬M#
#║## #║##
out
The "regulator" minecart sits on the NW track corner, spot marked 'M' to the right. A high-speed minecart enters the loop, collides with the regulator cart, the regulator cart goes around the loop and collides with the input minecart, pushing it off to the south.
The clincher is that the regulator cart is of a significantly different weight from the input cart. Heavier or lighter doesn't matter, only that there's a big difference in weight. An iron cart brakes an aluminium cart just as well as an aluminium cart brakes an iron cart.
You see, if carts of different weight collide, the output speed is reduced if a lighter cart bumps into a heavier one, but a heavy cart bumping into a light cart doesn't result in an increase in speed. So if you put two different-weight carts through a simple back-and-forth of collisions, the result is always a speed reduction.
Obviously, if both carts are of the same weight, the output speed will not be significantly reduced.
On the flip side, the whole thing is largely useless, because a derail-speed cart colliding with pretty much anything shotguns all of its contents. So this brake is only useful to reduce the speed of empty carts, unless you want to shotgun all your cargo into the speed regulator chamber for some reason.
PS: self-resetting design:
.#####
#╔╗╗#
#▲║║#
==╬=╝#
#║###
The track ramp is NW, so probably an impulse ramp or somesuch. This setup requires the regulator cart to be the lighter one, because it needs to be fast enough to derail over the ramp to bump the input cart. It's then re-accelerated by the ramp and returns to its starting position.
PPS:
Well, i actually started thinking about this when considering a way to sort incoming minecarts into automatic launch pads, in such a way that the cart would enter the first "free" location and pass by occupied ones. To be launch-able, the pad must be either a roller or hatch cover (or bridge, if you like your automatic launches via haphazard cart-slinging), so cannot directly contain a pressure plate. Of course, a return path with pressure plate can still be built, or a return mechanism rigged to get a cart back on the move if it unsuccessfully tries to enter an already-occupied pad.
Fast response, complicated:
. ║#### ║####
in==╚=║╗# in==<<▼^#
#╔╝║# #╔╝║#
#║╚╝# #║╚╝#
#║### #║####
To the left, the pathing, to the right:
<< = 2 long roller pushing west, rerouting incoming cart when pad occupied
▼ = 1x1 roller pushing south, actual launch roller
^ = pressure plate linked to activate the rerouting roller
An incoming cart will either find a vacant slot, goes through the loop and sorts itself into the launch spot. Or it will find the slot occupied, in which case it will bump the cart found forward, through the loop. The occupancy cart will pass over the pressure plate, activating the reroute roller and sending the incoming cart off through the reroute branch to the north, while the occupancy cart returns to the launch pad. The actual launcher needs an initial crooked path like here to allow a proper bump-loop onto itself. The reroute roller must be at least two tiles long - 1x1 rollers cardinally adjacent will power each other, so activating a 1x1 reroute roller would _also_ activate the launch roller. This design has the advantage of automatically redirecting additional incoming carts and should be safe from misfunctions when carts try to enter in quick succession.
Slower, easier:
. ║### ║###
in==╚=║# in=^<<▼#
#║# #║#
^ Pressure plate linked to _deactivate_ the reroute roller
<< reroute roller pushing west
▼ launch roller pushing south
This pretty much explains itself. The incoming cart deactivates the reroute roller upon arrival. If the slot is free, it passes over the inactive reroute roller and comes to rest in the launch slot. If the slot is occupied, the cart stops on the last tile of the reroute roller and stays there until the roller re-activates, sending it off to the next slot. If additional incoming carts arrive during the 100-step recovery period of the pressure plate, they'll expand the delay and may "clog up" the system by stacking themselves back until a cart permanently sits on the pressure plate and keeps the roller deactivated until you manually intervene and fix it.
If a level switch is acceptable, some rollers could be replaced with hatch covers (notably in the first case - instant reaction equals fast rerouting) or grates (second case - delayed opening of the grate agrees with the delay-dependent nature of that design).