Indexing can be permanently broken

When I have to re-index a site from scratch, I try to play with the number of nodes I can index per cron run. When I increase this number too high, cron dies with OOM. When I then decrease the setting again, cron will still die, because of the limit being set to first address the nodes that weren't flushed to disk the last time. However, this number of nodes can still be too high and so the only way to fix the indexing is to start from scratch with the lower setting.

I think that

if ($not_flushed > 1) {
// Re-index content that wasn't flushed to disk last time, minus the node
// which failed last time.
$sql = 'SELECT nid FROM {xapian_index_queue} WHERE status > 0 ORDER BY COALESCE(priority, 0) DESC, added ASC';
$limit = $not_flushed - 1;
}

should also take the variable_get('xapian_indexing_throttle', 100) into account.