A partial update:

  I am using 2.0.11 now.

  After updating to libgc 7.4.0, I have the same problem with a similar
  stack trace.  libgc is built without threads and passes it's "gmake
  check".

  I have the same problem with libgc 7.4.0 on netbsd-5/i386 (libgc
  passes gmake check on this combination also).

  On netbsd-5/i386, I tried to build libgc and guile with threads, but
  it failed to link with tls_get_addr.  This is likely a known lack of
  support in netbsd-5 which AFAIK doesn't provide thread-local storage.

  I haven't yet compiled libgc and guile with debugging symbols and
  without optimization; I realize that's what I need to do to figure
  this out.  I read a bit of the code to get started understanding the
  boot sequence.

  It seems there is detection of stack growth detection.  I wonder if
  that's somehow going astray.

I am curious if others are having success on other *BSD, and which
versions and architectures.