A few more points:

1. The Matlab tech support page says their double-precision tests don't 
show much speed-up for 64-bit over 32-bit; they recommend 64-bit primarily 
to allow larger array sizes.  However, 64-bit is only needed to address 
memory beyond 4GB, which we don't have in this case.

2. The Matlab Compiler will be calling the local GNU or Intel 
compilers.  If we go with the 32-bit version, then the buildopts.sh and 
mexopts.sh configuration files for the Compiler may need to be modified to 
ensure linkable 32-bit compilation or emulation.

3. The MathWorks tech support may be able to make some useful 
recommendations, both about multiple versions and the trade-offs of using 
32-bit over 64-bit on our particular architecture.

Steve