I was worried about this card. Cards that allow a good buy/gain in the same turn as heavy trashing tend to be powerhouses (Masquerade; Trading Post, when bought early; Forge, except that it costs $7). Pawn Shop allows this on an unprecedented scale: trash 3-4 Coppers and get a $3-$4 card too? Surely Pawn Shop needed to cost $5.
But I think it's just fine, mainly because unless you have help, you tend not to be able to use it much. After that first turn trashing 3-4 Coppers, you're lucky if you get to do that again. More likely you get to trash 1-2 Coppers a couple more times (so good $2 cards are important) and then it's a dead card. It's definitely a great early buy, but $3 is fine. It passes the "open with two" test that $3 cards must pass: since one Pawn Shop becomes a dead card quickly, two Pawn Shops become dead even more quickly.
Of course there is the other option -- breaking a card up into Coppers. That's a much more situational use, good with alt-VP (especially Gardens) and maybe once in the end-game as a last-ditch move to pull in an extra green card. As it's generally worse than Salvager (no +Buy, no Copper flooding), I'm not too worried about that one.
I did experiment with breaking up Estates into Coppers in the early game. On its own, that's probably a bad move, for the same reason that, generally speaking, you'd rather return two Coppers with an Ambassador than one Estate. But if those two Coppers can then be used by Pawn Shop to gain a better card later, maybe it's a good idea. My tests were too sketchy to be conclusive on this point, but it definitely did not play as sweetly as the idea perhaps sounds.
But all that changes with the right combos in place. I played Pawn Shop on a Tactician turn and trashed 6 Coppers for 1 Gold. Obviously that was amazing. I'm not worried about it -- it's a power combo that's fine being a power combo -- and there is a cost to delaying the use of Pawn Shop until you've got a combo (whatever it might be) set up in the first place.
Bottom line, this card is really nice, and after some cursory testing I can't find any flaw with it.