Third looks like a hard-coded, real thing. But, how to be sure it will be updated for your next SI version.
Imho it's really, really disappointing that, after all that years of ICE, still nothing of that basic stuff when it comes to curves, at least some "normalized U to location".
To explain it simple, if you want equivalent of XSI path constrain in ICE, you have to build a pretty complex ICE tree, all that not even close to performance of "ordinary" XSI path constrain.
At least from my little knowledge, when it comes to real life, a lot of curves with multiple
segments, let's say for hair.... only a very, very simple math is able to survive, if standard ICE trickery is used (select in array, modulo, so on). ICE Fitted bezier, Piecewise B Spline... this is somewhere at border of slowness, crashing, so on. Both even without lovely 'repeat' node.