In the creation of the Julian calendar, Julius increased the length of the months that weren't already 31 days long, except for February. Why?
The goal was to create a regular 365 day year that could regularly be reset every four years with a single intercalary day (there had been problems with the old system in that the intercalary month to reset the prior 355 day was not used regularly, and generally only by decree with not much warning - sometimes the decree wouldn't reach the provinces until well before the calendar transitioned from February to the intercalary, so records were out of whack). By 45 BC, the old calendar was 67 days adrift from the tropical year.
How do get a 365 day calendar with 12 months - the easiest thing is to set a minimum 30 days on each month and then add an extra day to 5 of those months.
Four months already had 31 days - and Caesar did not shorten any of those. But they weren't evenly spaced in the calendar (there's a 2 month gap between Quintilis and October, and 4 month gap between October and Martius). After 45 BC, there are no gaps longer than a month between the 31 day long months. This was achieved by adding not just one, but three long months (Ianuarius, Sextilis, December). So, 2 days had to be dropped elsewhere.
These 2 days could have been taken from 2 separate months - so the question is, why only Februaris?
It was already special. It was already shorter than all the months before the reform, and as T.E.D. noted, there seems to be a particular holiness about the month.
Since the Roman calendar used three "anchors" in each month (the Kalends [first day], the Ides [middle of the month], and Nones [8 days before the Ides]) and counted down to the next anchor, inserting an extra day in February would mean either the count to the Nones of Februaris or Kalends of Martius would increase - causing celebrations, festivals, and holy days to shift. The only problem with this reasoning is that Februaris was variable during years when the intercalary month was used before 45 BC, and remained variable in the Julian calendar to allow for the intercalary day.
Frankly, I think the decision came about from the following (in order of importance):
- 365 day year
- No month to be shorter than it was before
- No more than one month separation between long months
- Keep Februaris the same length (in regular years)
- Allow the inclusion of an intercalary day every 4 years
I would love to have the time to better reference this conclusion - but Wikipedia does offer a reasonable amount of detail, and their references are interesting followups (although, I can't read the ones in Latin). Ultimately, though, we may never know unless someone can dig up a parchment that Caesar scribbled any brainstorming on.