I was hoping there would be a quick fix for this, but the more i dug into the shard result merging logic for grouping, the less i understood it.
I went ahead and commited my tests, with a workaround to ignore the comparison of the missing group.query between the single node and distributed cases – so it still demonstrates that the distributed query doesn't fail, but it ignores the fact that the responses look diff.
I've opened SOLR-4181 to try and deal with it at a later date.
Committed revision 1421451.
Committed revision 1421456.