Multiprocessing on the PowerPC 7xx-series chips was difficult because, like the 603 they were based on, they lacked the proper cache coherency protocols and the like to support proper multiprocessing. It was possible to run a dual 603 (and Be, in fact, had one) and dual 750s, but it was extra effort required on the part of the system designer, and it wasn't really all that rewarding; it's like MPing with Celeroffs.
The PowerPC 7xxx series, however, like its 604 derivative, did have all of the necessary hardware for an MP design. Hence the reason these chips were used in multi-chip configurations from many different vendors, not just Apple and friends.
The memory/PCI controller used in the G3 line, the MPC106, could run in two different modes, according to Tundra/Freescale/Motorola's specs: it could either run a single processor with an external L2 cache, or run two processors without. It wouldn't go both ways; that's what the MPC107, the controller included with all of the G4s, was for.
As a side note, the MPC107 could only do dual processors, which may partially explain why nobody (AFAIK) has ever made a quad-processor G4 upgrade.
It's still pretty interesting that a dual 7400 would work at all on in an old G3. I'd be interested to know how they got it to do so, even if it was imperfect. Did anybody ever try one in a Yikes! G4?