Comparing the test output from JBT 3.1.0 builds:

http://hudson.qa.jboss.com/hudson/job/jbosstools-nightly-3.1.0.CR3/53/testReport/(root)/ (2264 tests, run on soa3-RHEL5-x86)
http://hudson.qa.jboss.com/hudson/job/jbosstools-nightly-3.1.0.CR3/56/testReport/(root)/ (2226 tests, run on on vmg18-rhel5-x86_64)

In both cases the SVN source revisions input to the build are the same; however, in build #53, 38 more tests are run than in build #56.

Similarly for JBDS:

http://hudson.qa.jboss.com/hudson/job/devstudio-nightly-3.0.0.CR3/45/testReport/(root)/ (2264 tests, run on hudson1-rhel5-x86_64)
http://hudson.qa.jboss.com/hudson/job/devstudio-nightly-3.0.0.CR3/46/testReport/(root)/ (2226 tests, run on hudson1-rhel5-x86_64)

Again, the SVN source revisions input to the build are the same; however, once again we have a discrepancy of 38 tests between builds.

–

The difference between these two in both cases is that the SeamValidatorsAllTests suite (38 tests) does not return any results because it times out before it can complete.

Can this test suite be broken up into smaller pieces so that it will run consistently on all hudson slaves? Note that it's apparently not a 32- vs. 64-bit issue because both #45 and #46 ran on the same slave node.