The execution resources for Intel's cores are such that they are frequently double what Jaguar can offer in terms of numerical counts, prior to realizing they can readily clock twice as fast, so the cores are fine. Other factors intrude, though in general the idea that a higher core count Jaguar was a sufficient or superior stand-in for AMD's desktop cores for the consoles says more about Bulldozer than it does anything else.
The L1 and L2 could thrash more readily in a multithreaded case, since they are per-core. This is mitigated and probably bettered for Intel by a far superior last-level cache situation compared to Jaguar, which is an area where there was a clear economy of effort for the console designers.
For Durango, we know from the leaks that the dual-module approach is particularly painful for remote accesses, as coherence between L2s is not an optimized case. The numbers are about as bad as main memory access elsewhere, which means main memory access for the consoles is probably incrementally more painful as a result.
The exact organization for Orbis is not detailed, although we haven't been given a reason to suspect otherwise. Vgleaks indicates extra care is needed to make sure not too much tries to cross between modules, which means the 8-core consoles shouldn't be treated the same as a PC 8-core. Assuming two reserved cores, a game faces four cores it can readily use and two that will tangle more frequently with the L2 traffic of the system reserve, and for which excessive interplay with the remaining four can cause performance problems.
There are still decoupled functions that can do well enough on those two, but that places a higher priority for the game on a single module.
For Intel, there's a very high-bandwidth and large L3 that is uniformly shareable. In cycle terms, it is (being clocked twice as fast aside) moderately longer-latency than the Jaguar's not particularly impressive L2 numbers.
