I can confirm seeing the same here - on the eu server.
Most of the time I have seen this, the cudaminer reports "GPU #0 12345678 hashes 200kh/s"
Then the miner seems to have no work for 5-60 seconds - GPU usage is 0%
It only becomes active again after "Stratum detected new block" is received.
No connection timeout or accepted/rejected is reported in this "pause".
Aha, then so far, three of us have the same problem. Perhaps others havent noticed this yet.
I also notice a few "Stratum connection timed out". It could be related.
Maybe the shorter periods also are "lost connections", but shorter than the miners "timeout" value?
A little more digging today.
When the GPU activity is zero, I also receive a "Get Failure" on cgminer.
It seems the worker cant get work, and pauses n*5 seconds to try get work again.
So this seems to be related to "Get Failures". And it is currently on both eu and us server.
(I have no statistics saying it is more frequent now than 1-2 weeks ago, though)