Shameless plug for the new version with "threaded" paths.
Here are some examples on de_aztec using threaded and non-threaded calls.
I used all the standard parameters of the natives. increasing StepSize will increase the performance. Heuristic will too, but will create a suboptimal path. (It will look like someone placed a magnet on the end position and dragged most points towards it.)
Non-threaded:
Code:
Start: -3030, 360, -325. End: 2480, -288, -277. Distance: 5548 units.
Execution time: 13s 533ms (88715 nodes, 46858 validated, 12902 successful)
Distance of path: 9248 units, 258 steps.
"Threaded":
Code:
Start: -3030, 360, -325. End: 2480, -288, -277. Distance: 5548 units.
Execution time: 14s 185ms (88715 nodes, 46858 validated, 12902 successful)
Distance of path: 9248 units, 258 steps.
Non-threaded:
Code:
Start: -3073, -925, -213. End: 2480, -288, -277. Distance: 5590 units.
Execution time: 1s 901ms (26951 nodes, 14084 validated, 4114 successful)
Distance of path: 7335 units, 221 steps.
"Threaded":
Code:
Start: -3073, -925, -213. End: 2480, -288, -277. Distance: 5590 units.
Execution time: 2s 115ms (26951 nodes, 14084 validated, 4114 successful)
Distance of path: 7335 units, 221 steps.
As you can see, there's a huge difference in just the starting position.