Hi Christian,
no, no need more test - Olav pointed out that MRR is used in your case because of the "read_rnd_buffer_size = 2M" setting (as the buffer is big, MRR has a lower cost), while I'm using a default value..
in any case, we all agree that for "LIMIT 1" case it looks like a bug, so should be fixed. (if you want you may log it yourself to be updated on changes (and just add a link to this forum thread)).
while for the initial issue with higher IO activity on 5.6 nothing is clear for me :-) so, if you can collect some stats with my own tools, then I'll probably say you more :-)
here is the howto:
-
http://dimitrik.free.fr/blog/archives/2012/09/mysql-performance-collecting-stats-from-your-workload.html
Rgds,
-Dimitri