There is a regression around 3% in jvm98 and 1% in scimark on Solaris X86 starting in 1.4.2b19. The Alacrity results are at the following:

http://alacrity.sfbay/query/nes.jsp?base=2468,2469,2470,2471,2472,2473,2484,2768,2772&build=9235,9236,9237,9238,9240,9241,9267,9268&pval=0.01

Actually, the regression was introduced several months back but has only been caught recently. The reason is that we recently changed running our Solaris X86 results on LX-50s from normal Dell X86 machines. Having rebaselined both the 1.4.1 FCS numbers and the 1.4.2b19 numbers, the regression can be seen only on the LX-50s. Doing a binary search on hotspot putbacks, I've narrowed the regression down to Azeem's c2_baseline merge located at

/net/balvenie/export/imgr_home/archive/main/baseline/2002/20020724102444.azeem.merge0724

Reading the comments of that merge, it seems like there are some x86 changes that might be affecting these results. The putback before does NOT have the regression.