This looks like some flakiness in Google Code. I'd suggest you retry the conversion, and report it to the Google Code folks if it keeps failing. As a work around, you can interrupt the clone after a few revisions, and do a pull from within the repository. This will allow you to simply resume the conversion if it fails.
I'm marking this as a duplicate of #225, another bug where I saw a similar failure when reproducing it. The problems later went away.
Anyway, I tried to reproduce this locally. Both using the regular SWIG bindings and the new Subvertpy bindings introduced in 1.2. Using SWIG, the clone failed at various points: after 538, 594, 2536, 1970 revisions, with the current clone being at r3600. The Subvertpy clone succeeded without a single failure. This all seems very odd to me, but I still consider gcode the most likely culprit.
Anyway, I'll upload a bundle of the successful conversion somewhere so you can use it.
It would be nice if you would upload it somewhere, thank you.
Pity for me: after 4 hours, 1.1.2 also crashed, at revision 9000. Now I've again at head version of hgsubversion but I've managed to interrupt the clone by Ctrl+C, so that at least something was placed in local directory, and now I do incremental pulls... Still running... Incredibly slow, this morning it was faster, so you seems to be right - it can be google code service quality.
Thanks for looking at it.
sorry for bothering you again - how exactly you did the convert?
The results seems to be very good.
I need the exact command line / used tools version to document the process in case the project will migrate to hg later.
Thank you again!