This is just an issue with display.  Instead of changing the actual pool server to spit back numbers and amounts of hashrate going to various coins (tougher to test, requires a server restart, etc), the chart of coin mining hashrates was just a quick little hack that looks at the last 1 minute of shares submitted in the database.  What that means is that once we switch, the stats look to lag behind a little bit, even though the hash power has completely switched.  10 seconds after a switch for example, the stats will show 50/60 seconds of the last minute's hashpower going to the coin we were previously mining, and 10/60 going to the one we're actually mining at 100% speed.  We could look over a shorter period without a problem, but the rates become far more skewed by variance.