101642 2013-04-09 16:29:09 1 20 1.0 2032890 30.0376 165.304 64.2496%
101641 2013-04-09 16:25:09 1 20 1.0 2032870 30.0351 165.301 64.2517%
101640 2013-04-09 16:25:09 1 20 1.0 2032850 30.0354 165.301 64.2517%
101639 2013-04-09 16:23:09 1 20 1.0 2032830 30.0343 165.3 64.2528%
101638 2013-04-09 16:22:00 1 20 2.113 2032810 30.0338 165.299 64.2534%
101637 2013-04-09 16:22:00 1 20 8.452 2032790 30.0341 165.299 64.2534%
101636 2013-04-09 16:22:00 4 5343.18432958 33.809 2032770 30.0344 165.299 64.2534%
101635 2013-04-09 16:22:00 1 20 135.239 2032750 30.0397 165.299 64.2474%
101634 2013-04-09 16:01:04 1 20 1352.383 2032730 30.0255 165.285 64.2585%
101633 2013-04-09 15:40:30 1 20 13523.835 2032710 30.0115 165.27 64.2695%
101632 2013-04-09 15:20:17 1 20 135238.105 2032690 29.9977 165.256 64.2802%
101631 2013-04-09 15:00:00 1 20 135238.105 2032670 29.9839 165.242 64.291%
101630 2013-04-09 15:00:00 1 20 135238.105 2032650 29.9842 165.242 64.291%
It appears the actual drop started on block 101633. Depending on the alog it may have taken a couple of blocks for the difficulty to actually drop, even though the version cutoff may have been on 101631.
For example noted in the code:
If the new block's timestamp is more than 20 minutes
// have to be greater than the max accepted time delta ; 15mins.
// this way, if one would artificially increase block nTime to its max value,
// we'd still take the 5mins periods without block before allowing a one-shot
// diff decrase, later keeping the block time used for ema computation.