How accurate is blockchain.info pool splits?

[DeathAndTaxes]

The reason I ask is because it looks like Bitcoin is becoming more decentralized.

http://blockchain.info/pools?timespan=14days

I remember when there was fear Deepbit alone would go >50%.  
Then it was Deepbit + Slush > 50%.  
Then Deepbit + Slush + BTC Guild >50%.
Now we are getting close to top 4 combined being <50%.

If p2pool (and smaller pools) could pull another 500 MH/s from the big 4 then even the top 5 combined wouldn't be 50%.  

[finway]

I think it's not accurate, cant' be there so much unknown hashpower

[bitcoinBull]

I was delighted to see p2pool on there and deepbit declining.  Also curious about their methodology and accuracy.