How to compare impact factor over time
If you want to see whether the impact factor of a journal is increasing, then you should look at how the same metric has changed over time. You could choose the 2 year or 5 year impact factor for various reasons. However, you need to focus on only one of these metrics at a time.
The scimagojr website has similar data that is publicly available based on slightly different sources to the ISI data, but it provides information on how indicators have changed from year to year. E.g., Cites per doc
Here's one example for a journal in my field: http://www.scimagojr.com/journalsearch.php?q=30073&tip=sid
What does a difference between 5 and 2 year impact factor mean
It's mostly caused by how citations to the journal are distributed over time. The citation half-life is a useful summary of this distribution. So, for example a citation half-life of 5 years means that half of all the citations that an article will obtain will be accrued 5 years after publication. The ratio of the 5 year to two year impact factor will be larger for journals with longer citation half-lives.
Importantly, fields vary dramatically in citation half-lives. Medicine, neuroscience, and nursing have short half lives, whereas psychology, mathematics, and the social sciences have longer half lives. So, using longer time windows for impact factor calculation will reduce differences between disciplines, although there are other metrics that can used if you wish to reduce discipline-specific bias.
Of course, it may also be that the underlying impact factor of the journal is increasing or decreasing over time, and this will also contribute (typically a small amount) to the nature of any difference between 5 and 2 year impact factor. Specifically, if the impact of the journal is increasing over time, the ratio of 5 to 2 year impact factor will be reduced.