First off, to answer the question, I gave my opinion that it is the GM's job to provide a game for the player characters he gets, not passive aggressively penalize the players for not picking an "optimized" party balanced to some kind of mechanical ideal. Control of the player's character, and definition thereof, is one of the few things that the players should have absolute control over (within parameters set by the GM, of course. I absolutely believe in the GM's ability to veto broken or "bad" mechanics.) It's not any more difficult or less fun for GM's to provide a game that works for the players rather than for an idealized theoretical party structure. In fact, except in very unusual cases, it's going to make the game much more fun for the GM to provide a game that's also fun for the players.
So, I reiterate; to sacrifice a fun game for one that's idealogically "pure" is not only bad GMing and a poor consolation prize, but it's also passive aggressive. I see no point in it.
But, that out of the way, I've actually, like I said, had the ambition of running an all rogue game for some time. The idea appealed to me when someone I knew through work mentioned years ago that they had done so with a pirate theme, but it didn't really congeal in my mind until the Paizo folks came up with the notion of the archetype--subtle variations to the character classes suite of abilities that optimize them for a more specific role or archetype within a campaign. Using the Pathfinder variant on the Rogue, including the archetypes introduced, has a number of benefits relative to the 3.5 SRD.
- d8 hit dice (as opposed to d6) removed some of the fragility associated with the rogue--which helps tremendously in an all rogue party.
- Pathfinder skill consolidation helps the skill points go even a little bit further. With 8 skill points per level, Pathfinder rogues can be built to be experts in pretty much anything, and a party of all rogues can contain an awful lot of varied expertise in a game that utilized a lot of skill checks. Which it should, especially given this party.
- 3.5 rogue is locked into a number of class abilities that may or may not be applicable to the concept of the character. This is why the archetypes are so nice; the Trapfinding abilities in particular get frequently sacrificed to make way for something else (since Trapfinding is a very specific ability that doesn't apply to a lot of Roguish archetypes.)
- The pickable abilities are greatly expanded, making room for additional flexibility and variability in rogue builds.
From the Advanced Player's Guide section of the PRD:
- Roof Runner
- Sactified Rogue