[lmi] 64- vs 32-bit msw performance

[Greg Chicares]

Time (ms) to run a composite for [unpublished] test/ten.cns (153 cells):
  32-bit msw:  12529  9336  9797  11954  9508
  64-bit msw:   7517  7227  8058  10370  7349

9797 / 7517 median  run is 30.3% faster
9336 / 7227 fastest run is 29.2% faster

I didn't specify any non-default floating-point options, so I guess
the 64-bit build is using SSE. I haven't tested its accuracy at all.
Much work remains to be done, but this does seem encouraging.