We're seeing a major performance regression with ECperf on 1.4.1 build 8 and later.

Here's some data taken from the Forte Analyzer that illustrates the problem.

  %time in 1.4.1 %time in 1.4.0
frequency_counter_overflow 10.3 .1
   method_invocation_event 2.2 0
   method_back_branch_event 5.6 0
      compile_method 6.8 0
         resolve_string_constants_impl 3.8 0

Times are % times of a 30 minute run of ECperf and include the children methods shown.

This is not completely deterministic; we see the regression in about 90% of our runs.