The most important component in RPI is a team's record (even though it's only 25% of RPI, it's where the most variation occurs), so a team like Princeton is going to have a good RPI ranking. Princeton also played a pretty decent OOC schedule, with 2 RPI top 50 opponents and 5 more RPI top 100 opponents, but its RPI probably actually will go down the rest of the way because the Ivy League isn't very good overall.
South Florida (which dropped out of the top 25 today, by the way) is benefitting from a strong OOC schedule (4 RPI top 25 opponents) and a generally good record. The Bulls have only 1 RPI top 50 opponent left, though (UConn), and so there's a decent chance they will drop, too.
The committee has said for years that RPI by itself is not really a factor in seeding, but that they use it to look at how good a team's wins are and how bad its losses are - RPI top 50 and 25 wins (and to a lesser extent, RPI top 100 wins) are good, and sub-RPI 100 losses are bad.