Very slow benchmarks (and in practice) -- how to begin diagnosing the issue?

I've got access to servers on a cluster through my research group, and we've got a very competent sysadmin who I'm trying to help diagnose the issue. Basically, matlab is slow. Very slow. Here are some benchmarks I ran on a few different servers (these are all enterprise quality running Linux 64-bit; not sure about the chip architecture; tons of RAM and multicore.)

matlab -nodesktop -nosplash -r "bench(5)"

ans =

    0.3552    0.1913    0.2610    1.9885    0.4658    0.5240
    0.2382    0.1751    0.2433    1.8276    0.4809    0.4884
    0.8841    0.2238    0.2437   13.1059    0.5288    0.5200
    0.7138    0.1989    0.2481   10.1533    0.6426    0.5555
    0.4868    0.2106    0.2508   12.7071    0.5702    0.6880

matlab -nodesktop -nosplash -r "bench(5)"

ans =

    0.5810    0.1906    0.3687    7.0276    0.5580    0.5481
    0.4653    0.1847    0.3011    6.7573    0.9881    0.5759
    0.5204    0.2062    0.2441    8.1615    0.6202    0.4723
    0.5560    0.2074    0.2785    6.8986    0.5370    0.5196
    0.5753    0.3601    0.2824    6.9958    0.4643    0.6548

>> bench

    0.6941    0.3787    0.2958   14.4527    0.5514    0.5435

>> bench

    1.0360    0.5322    0.2471   16.4350    0.5510    0.6338

I found another benchmarking script here: http://math.ucdenver.edu/~jmandel/matlab/matlab_benchmark.html

Server 1:

1. large LU:   0.860
2. small LU:   0.944
3. sparse  :   0.302

Server 2:

1. large LU:   1.813
2. small LU:   0.526
3. sparse  :   2.216