The statement does indeed hold in type H3; you can find the computation at http://sage.lacim.uqam.ca/home/pub/15/. The machine is still running (and I don't know how long it might take) doing type H4.
So, if no Coxeter theoretic proof is known, there is hope to find one...