That paper isn't particularly relevant, as the non-OoO CPU described therein is not much like the Mill at all. It would be interesting indeed, though, to see the Mill folks produce performance figures for some common software.

Remember that a VLIW CPU has a lot of sub-ops in each instruction word, so it by design has more ILP, though only when the workload allows for many of those sub-ops to be filled, and how much work those sub-ops do can muddy the waters.