This is the same code submitted 4 times:
The first submission spends 2137ms on the 3rd testcase while the third exceeds 3000ms.
Moreover I added this problem to my mashup to set custom TL. One submission spends 3432ms, another 3790ms.
So the difference is more than 1000ms.
The reason might be huge bitsets containing 300000000 elements which I use in my solution, but still I think that it isn't a correct behavior of Codeforces judge.
Does anybody know what causes this instability?