ok, bear with me here. I'm going to start with the utmost basics and build up, so as to make the reasoning for starting this discussion clear. I'd like to make two factors known in advance, first is that I have no opinion as of yet, second is that just because something is currently done one way (as in Purge) doesn't mean it's neccecarily the best way. I trust no one will cite their difference in rarity as an issue with any seriousness (cremate is an uncommon).
Right. Now, for clarity's sake:
Coffin Purge

Instant
Remove target card in a graveyard from the game.
Flashback

Cremate

Instant
Remove target card in a graveyard from the game.
Draw a card.
The most obvious thing about these cards is how similar they are. Both are instants and have all the benefits that implies, including synergy with Cunning Wish. Slightly less obvious is they rely on cunning wish to be useful at all, as no compeditive deck has the amount of free sideboard slots, let alone cards to side out, that running these things as conventional sidebard hate would require. You're going to have one or two, and you're going to use Cunning Wish to grab it if you need it. Because this reduces the number of decks the cards are really eligible in (ones that maindeck multiple cunning wish) there are less randomness factors to consider when comparing the cards.
Another readily apparent similarity is that both cards, once cast, replace themselves with another available spell, either by cantriping or by effectively being still in your hand via flashback. Herein lies the difference though: one spell puts a copy of itself in your hand (which is visible to your opponent), the other spell puts a random topdeck in your hand (which is not). The question becomes which is more useful to you.
Consider context. First of all, you are not wishing for the card unless you already need it. Chances are, unless you wished it out to rfg a regrowth target or randomly stop 'tog from reaching lethal damage after their hand is gone, it will be useful again, whether you are dealing with a goblin welder or a worldgorger dragon. However, unless you're playing a bad deck or very poor metamage choice, the next card in your library stands a good chance to be useful shortly as well.
Consider card slots. If you rfg one problem card and may need to do so again, you can run just one coffin purge and buy yourself two chances to stop something very bad from happening, and by extention, need only burn a single Wish for those two chances.. This is a serious consideration against the current recursion combos, which are extremely redundant.
Consider efficiency. While it's good to have an ohshit button should your deck fail to stop a Bad Thing, it's often stronger still to stop the real problem cards. Obviously, removing a welder is a much better use of a card than removing a mindslaver or platinum angel. Dragon is a bit of a different story however, the enormous redundancy of recursion spells in Dragon decks makes countering them all unlikely. Yet in either case, Cremate provides a fair chance at doing something productive right now instead of buying you time to do something productive later should a second Bad Thing come about.
Consider dynamics. If you stop one threat, are you likely to have need of another casting of that same spell before the game is already decided? If not, the new spell is probably better. With the amount of "swingy" spells in type one, crushing Dragon's attempt to go off, and to a lesser extent Slaver's attempt to combo out, should mean you have a massive, full-turn-minimum window of opportunity to kick your opponent's teeth in. Yet this is hardly a sure thing. Using this same logic, is stopping TWO such attempts not an even bigger window of opportunity? Or is it just win-more? Does the random topdeck from Cremate significantly improve your next turn to the point where having the extra card gives you more to do during your window of opportunity (it COULD be yawgwill... or a fetchland)?
Don't underestimate the cantrip, if it's tied to a useful effect it becomes extremely powerful (time walk and an active future sight are the best examples). That's not to say you should overlook the surefire redundancy of a two for one.
Thoughts? As I said, my jury's still out on this one.