When reading my textbook I noticed that in all examples but one from eight the recognition site was an even number of bases.

I wondered if this was just a co-incidence, so I took the data from this site for over a thousand known recognition sites and put it into a spreadsheet (XLS uploaded here). The results are probably best summarised graphically:

Can anyone explain or suggest why it is more common that restriction enzymes recognition sites have an even number of bases?