...
I've built a dc dummy load and used a couple of lm317's in parallel in cc-mode combined to draw 1A, a useful number for diy sized battery's.
That is good from 5 Volts up to about 15 Volts (15 Watts dissipated) above that I can connect regular 12 Ohm (50 Watt) power resistors in series for higher voltage's.
Accuracy from those lm317's is pretty good with time and temperature, deviation on mine is below 0,1 % 
These are on a good sized heat-sink, keeping temps low helps.
I did something like this once for a factory test jig, for telecom power supplies. The jig provided a programmable constant-current load for the UUT, 5V to 12V, 20A. Ended up using a number of 3-terminal regulators each sinking a fixed current, plus a power FET in a moderately slow control loop, all in parallel.
When loading the UUT, the jig would switch on the 3-terminal regulators (one at a time or all at once, depending on the need) to sink nearly all of the required current. The jig would then enable the FET circuit. The FET circuit would "clean up" any error, between the programmed setpoint current and the current sensed through a 4-wire resistor made of bare manganin wire. So it would compensate for thermal drift in load resistors, in regulator characteristics, etc.
When removing the load, the jig would do the reverse sequence: variable FET load off, then the fixed loads off. Worked pretty well. The regulators and load resistors were cheap, and I got a metre of manganin wire as a sample. The jig had lots of heatsinking, and a careful placement of fans in a push-pull arrangement for forced air cooling.