1) make a note of the server seed hash for this roll
2) pick a client seed in an unpredictable fashion
3) make the bet, noting the number they rolled
4) verify that the hash of the published server seed matches the one in step 1
5) verify that the published server seed combined with my chosen client seed gives the roll I saw in step 3
That's a LOT of work. Almost nobody is going to bother doing it, and so for all practical intents and purposes the site isn't provably fair for the majority of people.
The argument I'd make is that this longer process allows for real time verification, which I find preferable. People are either going to take advantage of provably fair functionality or they're not, and most of them are not. Those who are taking advantage are either going to only make a few rolls and be done, so it hardly matters if there's a few extra steps involved, and I'd argue that it's unlikely they would return to see the server secret after 24 hours anyway, or they're going to automate the process, so again it doesn't matter if the process is longer or not.
The major problem I have with a daily secret type setup is that it exposes the site operator to risk should that list ever get out. I agree that JD's system is a decent compromise, as at least bets can be verified every 10 rolls or so as opposed to every 24 hours.