Mircea, is there part of resource ordering that might lead to a large amount of disk activity?

I do not believe there is. Resource ordering is reading a few resource-dependency.xml files that is all. They are already available in the classpath.

Test 1

trunk non-EE - loaded first
Time taken by phase id --> RESTORE_VIEW 1 in MS. 92
Time taken by phase id --> RENDER_RESPONSE 6 in MS. 3515

trunk EE - loaded second
Time taken by phase id --> RESTORE_VIEW 1 in MS. 74
Time taken by phase id --> RENDER_RESPONSE 6 in MS. 2225
----------
Test 2

trunk EE - loaded first
Time taken by phase id --> RESTORE_VIEW 1 in MS. 69
Time taken by phase id --> RENDER_RESPONSE 6 in MS. 2554

trunk non-EE - loaded second
Time taken by phase id --> RESTORE_VIEW 1 in MS. 63
Time taken by phase id --> RENDER_RESPONSE 6 in MS. 1905
----------
Test 3

trunk non-EE - loaded first
Time taken by phase id --> RESTORE_VIEW 1 in MS. 70
Time taken by phase id --> RENDER_RESPONSE 6 in MS. 2091

trunk EE - loaded second
Time taken by phase id --> RESTORE_VIEW 1 in MS. 68
Time taken by phase id --> RENDER_RESPONSE 6 in MS. 1871
----------
Test 4

trunk EE - loaded first
Time taken by phase id --> RESTORE_VIEW 1 in MS. 69
Time taken by phase id --> RENDER_RESPONSE 6 in MS. 2190

trunk non-EE - loaded second
Time taken by phase id --> RESTORE_VIEW 1 in MS. 84
Time taken by phase id --> RENDER_RESPONSE 6 in MS. 1953
-----------

As you can see from the results above there is no difference for the execution time between the EE and non-EE versions. The only thing that seems to matter is the order in which the applications are accessed. Every time the application accessed first will run ~150ms slower then the second one. Most probably difference is due to classloading, the second app class loader does not need to load so many classes and resources anymore.

The application server was restarted before each test was executed.

Also ran YourKit profiler with both applications and could not detect any significant differences between them. The profiling results show that the majority of CPU time is used for loading the resources (the mandatory ones).