Hey Frank, Hey Wyatt!
Yeah, the disk-brake works fine, as long as you're not doing long-term loads... the disk heat-soaks quickly, and with cooling time between pulls being relatively short, the disk doesn't last too long. Personally, I wouldn't use a screw to actuate the master cylinder... I'd probably just make a special lever to pull by hand, mebbie a little longer than the original, and a master cylinder with a little smaller diameter than normal, to decrease effort... simply 'cause it's faster between application and release. Getting a 'steady' application isn't really all that tough, and a 'pull' will only last about 3 seconds.
Wyatt's tone-wheel is a good suggestion, but if you're doing pulls at 5000rpm, and it's a 100-tooth wheel... that's a pretty high frequency tone... my choice would be to use a hall-effect or active-reluctance (amplified-coil) sensor of some type (probably an industrial device scavenged from one of my clients' junque-piles).
As for interfacing, my personal method would be to make a circuit specific for gathering data- probably use a MicroElectronics PIC 16F876 at 20mhz to count tooth pulses and measure reaction-arm load. If I was really concerned about accuracy, I'd probably use a made-for-force type load-cell that falls in the de-facto industry standard of 12-15vdc in, and either 2-20mA out or 2-12v out pressure signals. I'd have the PIC just grab a series of pulse counts and ADC conversion counts, and dump 'em out the PIC's serial port to a PC, and make a silly program that just grabs it from the serial port, calculates RPM, converts the ADC counts, and poops it into either a text-file or spreadsheet.
Without doing a PIC-type sampling adapter, a good software-type guru could use the PC sound-card, with one input going to the speed sensor, and the other going to the load-cell... most sound-cards will tolerate and measure a certain amount of DC offset, so you'd probably be able to get a decent reading by simply writing your own DSP code to grab and process... or you could look in Digi-Key, Mouser, or other places for a plug-and-play type compu-scope adapter... they're awfuly pricy. For comparison, the little PIC-based telemetry adapters I make for one of my clients run about $20 a unit in quantities of 250, including all parts, labor, etc... the dyno adapter wouldn't require any more parts or complexity, so it'd probably be in the same realm... just a matter wether the world of Average Joes really wanted their own dyno.
As for software (and I haven't done much programming since using PASCAL in college)
I'd leave the rpm/torque relationship to a spreadsheet algorithm... as I'd want to compare both power-at-a-given-point, AND integral power, as well as the effect of engine inertial moment... not something that's necessary for the average tractor engine, but if I hooked it to a go-kart or motorcycle engine, I'd want to know how the engine's moment would affect my time-through-gears... it obviously has much higher effect in lower gears, but not as much in higher... so being able to measure and predict makes it easier to determine where to draw the line between smoothness and rip-snort zippy.
But I'd hafta build mine using about a 100gpm/4000psi hydraulic pump and a BIG heat exchanger... and a jackshaft that'll plug into the propshaft of my boats... something that'll soak mebbie... 350-400hp...