I'm pretty sure that the 250KB limit has never been broken to date.
block 187901 499,245 bytes
block 191652 499,254 bytes
block 182862 499,261 bytes
block 192961 499,262 bytes
block 194270 499,273 bytes
These are the biggest 5 blocks up to the checkpoint at block 216,116. Interesting that they're all 499,2xx bytes long, as if whoever is mining them doesn't want to get too close to 500,000 bytes long.
I understand that at least one miner has their own soft-limit, probably Eligius and probably at 500kb.
I take that back because these blocks are version 1 and Eligius is supposedly producing all version 2 (
http://blockorigin.pfoe.be/top.php)
However, I have extracted the transaction counts and they average 1190 each. Looking at a bunch of blocks maxing out at 250Kb they are in the region of 600 transactions each, which is to be expected. Obviously, there is a block header overhead in all of them. But this does mean that the 1Mb blocks will be saturated when they carry about 2400 transactions. This ignores the fact that some blocks are virtually empty as a few miners seem not to care about including many transactions.
So 2400 transactions per block * 144 blocks per day = 345,600 transactions per day or
Bitcoin's maximum sustained throughput is just 4 transactions per second. This is even more anemic than the oft-quoted 7 tps!
Excellent. Someone finally quoted some real numbers instead of theoretical maximums, the picture is a bit clearer now, thanks!