Looks like they posted some new demos - I love the chandelier - that's pretty hard core joints.http://www.numerion-software.com
I wonder about these two quotes, to me they sort of undermine each other:
"The mass of each box is the same - irrespective of its size."
"This is a demo you can't do with other popular physics engines"
I've never used Havok but in our own (2d) solver we've made a similar structure (a 5-level "mobile") that's stable with very few iterations; I would imagine that proper physics solvers can handle this ideal case too -- a tree (i.e non-cyclical network) of spherical joints with unit-mass bodies and no joint limits is a pretty ideal/friendly scene.
I'd love to see what happens when someone yanks really hard on one of the leaf nodes in that demo or tries to pull opposite leaves in opposing directions.
The truck demo on the other hand seems VERY impressive: 27k soft tetras constrained to rigid cores supporting a heavy load.
It's definitely quite interesting, it would be nice to have some more details about their solver; their talk of "discrete nodal" modeling (I read: particles) and higher-order constraints (I read: tetrahedral volume constraints) reminds me of Stam's Nucleus simulator.