For what sort of workload can a Core i7-920 be faster than a i7-860?

Question

Core i7-860 has a slightly higher clock speed (2.8 vs 2.66 GHz)

Core i7-920 has QPI and tri-channel DDR3 (vs DMI and dual-channel)

Answer 1

AnandTech have a good comparison of the performance of these CPUs. The performance is very similar, but overall the i7 860 is faster. The i7 860 is a newer CPU, but there are some benefits for opting for a Bloomfield CPU (such as the i7 920):

1) High-end multi-GPU performance (or other uses of high bandwidth PCIe)
2) Stock Voltage Overclocking
3) Future support for 6-core Gulftown CPUs

As you point the i7 860 lacks the QPI, and so memory-intensive applications should tend to be faster on the i7 920:

There are definitely cases where Bloomfield's memory controller is a boon, the Core i7 860 is able to approach but not outperform the i7 920.

Edit: Using virtual machines and (extreme) gaming are the only things I can think of that the 920 would be better at, but decent benchmarks are needed to confirm this hunch (also the hard drive may end up being the bottleneck for VMs). This site has an excellent dissection of the pros and cons, and comes to more or less the same conclusion. And this post reinforces that the 920 has greater PCI-e potential than the 860 - which may be important if you plan to opt for dual graphics cards for gaming/graphics performance:

1. P55 motherboards are less expensive than X58 motherboards.
2. P55 motherboards have no PCI-e slots free to use with raid controllers or other cards.
3. Overclock capabilities at stock voltages is very limited with the i7-860.
4. Support for QPI has been abolished with the i7-860.
5. P55 [860] is limited to 16 PCI-e lanes, X58 [920] has 36 lanes.
6. X58 is fully compatible with the new hex-core Gulftown, P55 is not.
7. P55 is limited to 4 memory slots, X58 has 6 memory slots.

Answer 2

The 920 will be faster when your problem is memory bound and not compute bound. So most things with large working datasets will be faster with it.