> Were the 24k shares being offered by one seller/broker, as in "I have 24k shares to sell at 17" or was the 24k just an aggregation of the availability all the smaller offers?

The shares were being quoted on different exchanges, at the same ask price. 24k was the cumulative volume that the buyer wanted, but that couldn't be fulfilled by a single exchange (the quote was for a smaller volume at that price). Therefore, to buy 24k shares, the buyer needs to trade twice, once at each exchange.

> I wonder why these trades are not being performed in parallel across the various exchanges, partially preventing this kind of arbitrage?

As has been pointed out, this is what a good broker will do - they will compensate for latency to make sure that bids arrive at differing venues at the same time to prevent the market shifting underneath them. A naïve broker will simply send out the bids at the same, and latency means that they arrive at different exchanges at different times. This lets the sellers at the more distant exchange move the market in response to the information of the trade being executed at the closer exchange.