I understand your graph now. Thanks Bill!
Juan, apologies for the rather brief response earlier. I had arranged to go for a long walk with some friends, and only had time to fire off a few quick words before I had to go out.
I had actually been a bit lazy with my choice of dataset. The .csv file available from the ADS Version 1 page is already laid out in a rectangular format, with the day/month down the vertical y-axis and the year along the horizontal x-axis. As that layout is ideal for this type of analysis, it was something of a no-brainer to use that file. A quick perusal of the full dataset available on Version 2 had already convinced me that there was no point in expanding the analysis, as there were no serious contenders amongst earlier years for positions amongst the "lowest 3".
As I'm sure you know, Version 1 does contain data for 2002 and 2003, but these were deliberately omitted. The data for 2002 is incomplete, and there were no values for 2003 which fell into the "lowest 3" category.
The Version 2 graph which you included does actually highlight a rather bizarre artefact of the data. Although 2005 had about 286 dates which - at the time - were the lowest on record, only 3 of these remain within the current "lowest 3". On the other hand, by the time that all the 2005 data had come in, 2004 was left with just 61 dates which were - temporarily - the lowest on record. At the time of writing, 2004 no longer has any lowest values, but still retains 11 second and 8 third lowest values.
If you look carefully at the chart which you posted, you can see that 2004 is lower than 2005 during an approximately 6 week period, stretching from mid-April until late May. All bar one of the still current 2004 "lowest 3" dates occur during this window. (The other was 29th Feb)