> FTR, I haven't checked this on Ubuntu 32-bit. It's rather odd for the

> failure to be architecture-dependent. Are these two machines running
> different versions of Ubuntu, perhaps?

They are (Ubuntu Natty 32-bit failing, Ubuntu Oneiric 64-bit passing).
However, I don't believe the failure is directly architecture-related, but
indirectly so. For example, test execution order might not be identical
on all machines, depending for example on hash values and dict bucket
sizes, which I would expect to be different on different machines for any
number of reasons. In my limited testing, I found the following:

Ubuntu Natty 32-bit - failed.
Windows 7 32-bit - failed.
Ubuntu Oneiric 64-bit - passed.
Linux Mint "Katya" 11 64-bit (based on Ubuntu Natty) - passed.

With non-deterministic test execution order combined with module-level
data updated by tests, it can be hard to reproduce these failures - a bit
like uninitialised data errors in C :-(