I'd say, based on the current level of PS2 emulation on PCs, that the prospects aren't great. I also don't have high expectations for Gamecube to be that great on high end phones today and I don't have high expectations for Pandora 2 to exceed those phones (actually there's a good chance it'll be weaker instead).
3DS isn't even worth speculating over until emulators are relatively mature on PC. Right now they haven't even been started.
In terms of hardware, how similar is DS to 3DS?
Going off Wikipedia and the associated suspected stats:
Nintendo DS:
ARM9 67MHz & ARM7 33MHz
4MB RAM
Nintendo DSi:
ARM9 133MHz (2X) & ARM7 33MHz
16MB RAM (4X)
Nintendo 3DS:
ARM11 MP2 400MHz (Common WAG) (~8X)
128MB of RAM (32X)
PICA200 GPU (Much more powerful then the DS)
The clock speed of the ARM11s in the 3DS is unknown. The clock rate does appear to be higher then the DSi, and the ARM11 architecture should result in a higher average operations per clock then the DSi hardware.
From what I've seen a Rebirth unit with a 850MHz Cortex A8 does okay, but not wonderful with emulating a DS. Now ARMH lists the Cortex A8 as having a theoretical 2.0 DMIPS/MHz, and the Cortex A9 is rated at 2.5. Newer processors don't appear to have this rating, but the rumor mill put the Cortex A15 at around 3.5 DMIPS/MHz.
At a conservative but reasonable 1.5GHz the Cortex A15 should have 162.5% more performance then the 1GHz Pandora model. At a unreasonable 2.5GHz, the Cortex A15 should have 337.5% more performance then the 1GHz Pandora model. Multicore versus single core is not being accounted for in that, however if the 3DS is running both ARM11s at, at least 266MHz, based on the current implimentation you're probably asking too much from a potential Pandora 2 to be able to emulate it based on the current approach. If it's closer to the 400MHz estimates, back of the napkin ballparking points to a need for a more efficient approach then current.
If they do decide to refresh associated with the 28nm node next year or the year after with a 3DSi, it's entirely possible they'll switch those ARM11s for Cortex A5s and bump the max clock speed up. Not so much in the interest of increasing performance as a Cortex A5 should have less die area at 28nm then an ARM11 does at 45nm, and battery life should be improved with the newer architecture and process.