We demonstrate the feasibility of our approach to reduce the solution of
vector-
valued problems from CSM and CFD to sequences of scalar problems to be
treated
with optimised and architecture-aware scalar parallel multigrid solvers.
For
several prototypical applications from linearised elasticity and fluid
dynamics,
the approach maintains weak scalability and node performance of the
prototypical
Poisson problem. In particular, FEAST applications on NEC SX-8 execute
significantly faster than on commodity based clusters.