Severe bug in GHC HEAD

[Kyra]

Ghc HEAD has some severe bug. A have not filled an "official" bug report 
since I nave not isolated it properly. Below is how can I describe this bug:

1. Symptoms: under a heavy load *and* with RTS option -Nn where n > 1 my 
application stops responding to (http) requests. With -N1 all is ok.
2. Hardware: 4-core Intel CPU, OS: Windows 7 x64
3. ghc-7.1.20110201- are not affected, ghc-7.1.20110202+ are affected, 
stable ghcs are not affected too. So, the bug was introduced in a HEAD 
branch between Feb 1, 2011 and Feb 2, 2011.
4. If I increase "n" in -An RTS option my application proccesses more 
requests before choking. So one could guess the bug is GC-related.