I tried an other approach: I loop with getVersionHistory() which gives me oldid, date and size which I need, and call getOldVersion() for each oldid. The sad experience is that the script becomes extremely slow. As the page under investigation is usually quite big (it's a long talk page) and the loop is so slow, I got the idea that there might be a memory problem with the first version.

Is it a possible behaviour of Python 2.7.2 that instead of explicitely throwing a memory error it just breaks from the loop without notice?