With permgen removal, all of the names of types and functions are going to change. The ordered mapfile symbols will be obsolete. The script and method of calculating the optimal symbol ordering has bit rotten. Maybe it can be fixed. It's attached.

This ordering may provide startup benefits on solaris machines. On an old solaris system, running with old refworkload version 82.

philli% environment
date: Wed Jun 13 18:48:48 EDT 2012
uname: SunOS philli 5.10 Generic_118833-36 sun4u sparc SUNW,Sun-Blade-2500
uptime: 6:48pm up 135 day(s), 57 min(s), 2 users, load average: 0.04, 0.06, 0.25
model: SUNW,Sun-Blade-2500
memory: Memory size: 2048 Megabytes
I-Cache in Hex: 00008000
D-Cache in Hex: 00010000
swap: total: 2640384k bytes allocated + 26912k reserved = 2667296k used, 7116784k available
E-Cache: SUNW,UltraSPARC-IIIiMB
cpu: 2 at 1280 MHz

Results:

logs.client.map
  Benchmark Samples Mean Stdev %Diff P Significant
  startup3 20 4.07 0.03 -0.25 0.626 *
  Framer 20 0.47 0.11 1.48 0.865 *
  JEdit 20 2.69 0.03 -1.55 0.227 *
  LimeWire 20 4.89 0.05 0.14 0.707 *
  NetBeans 20 11.46 0.07 0.29 0.292 *
  Noop 20 0.19 0.01 -2.71 0.008 Yes
  XFramer 20 0.50 0.03 6.48 0.330 *
============================================================================
Results with removing the -xF option:
============================================================================
logs.client.map
  Benchmark Samples Mean Stdev %Diff P Significant
  startup3 20 4.03 0.12 0.27 0.735 *
  Framer 20 0.49 0.16 -5.66 0.549 *
  JEdit 20 2.59 0.22 1.89 0.407 *
  LimeWire 20 4.86 0.06 -0.08 0.814 *
  NetBeans 20 11.47 0.09 -0.20 0.414 *
  Noop 20 0.19 0.01 -0.81 0.458 *
  XFramer 20 0.50 0.03 0.80 0.646 *
============================================================================