Bit of a funny one this one - there appears to be from what I have gathered a potential bug within the linux kernel that timestamps of memory mapped IO files do not get updated when data is written to the file ...
http://www.mail-archive.com/rrd-users@lists.oetiker.ch/msg14483.html
http://www.mail-archive.com/rrd-users@lists.oetiker.ch/msg14481.html...