There are some great QT documents from the Scalability Center that will speak to this - From my experience the best server performance is to have all slots in the chassis loaded with processors, rather than filling only two slots of a four slot machine. Optionally the customer could get the two slot Dell with two processors, but that would potentially limit the capacity of that server. The other way would be to load all four slots with less capable processors i.e. two core vs four core, then as the requirements grow they would just swap out the smaller processor for larger quad core or hex core processors. I'll have a look for the documents and post those in a minute.
you get a bigger overhead in data and or response time on the hardware, for each component which need to be address.
Between a 2 socket and 4 socket motherboard you have, if the design is optimized, more complexity in the inter socket connect. This mean the communication between the cpu.
You can find some information in the atachment.
Both the R910 and the R720 (fully stocked) are great selections on hardware.
There are some key differences though, particularly with supported chipset and CPU interconnects (QPI links).
The short answer is that performance would be improved with the R720 over the R910 with only two chipsets.
Especially if a fast CPU is used for R720 (such as the E5-2690).
The R910 only has one QPI between the CPUs whereas the R720 has two QPIs.
Add the increased clock frequency for the R720 and the selection should be straightforward.
Generally Publisher tasks (reloads) have been shown to have a high correlation towards clock frequency and a strong recommendation would be to go for the R720.
Comparing a fully stocked R910 versus an R720 as a QlikView server is harder as the R910 regains performance by having much more computational capacity despite the slightly slower clock frequency.
Hampus von Post