It has to do with the efficiency curve of the D5 pump. It's most efficient between 2 and 3.5GPM...in fact, over that range, it's 50+% more efficient than it is at 1GPM. At face value that doesn't mean much, but it's the underlying key to why the Typhoon III ends up being such a step forward in design
What does that mean for the end user? Well, let's take a sample loop and play with it
Sample loop: a D5, two MCR320s, an HK3.0, and two MCW-60s. With this loop in serial, you get roughly 1.4GPM (at 4.1PSI of pressure drop)....which is good, but maybe a little lower than you want. Because the D5 is only running at 1.4GPM, it's efficiency is under 14%. The overall pumping power the water is being subjected to is 2.5W.
Add in the Typhoon III and you can arrange your loop (by default) in a parallel config....you split the restriction up evenly: an MCR320 and the HK in one portion, and an MCR320 and the two MCW60s in the other. Now you have a few things going on (mathematically) that I may not be describing 100% clearly or accurately, but hopefully I'll get the gist of it across.
1) When you go in parallel, the flowrate through the pump is the sum of the two subloops.
2) When you go in parallel, roughly the average of the pressure drops of the subloops is the pressure drop across the entire loop (or across the pump, depends how you look at it). This means that at any given pumping power, flow increases.
3) Because restriction is down overall (by a lot), flowrate increases. When flowrate increases to the 2.0 to 3.5GPM range (from the <1.5GPM range), efficiency increases noticeably...so you actually have more pumping power at work.
You throw all three of those into an equilibrium (such is fluid dynamics....the three variables have a lot of dependence and codependence) and you actually end up getting higher flowrates in each subloop than you would in one big loop! Back of the napkin math says roughly 1.5GPM through both loops
Loops that are naturally more restrictive (say 1.2GPM or lower in serial) benefit even more!
A 'fun' loop to demo this on could actually be something like the HK 3.0 + a pair of GTX rads + a pair of GPU blocks + a motherboard block. Whereas you'd get maybe 1GPM when arranged all in serial....you can put the HK in its own subloop (it's the only thing that cares about flow) and you'll end up getting really high flowrates through the HK and still get moderate flowrates through the everything-else loop.
Because of the nature of what we do (how everything we do tends toward an equilibrium), you'll see no thermal issues from even the wackiest of the loop divisions....if you want a semi-restrictive CPU block on its own subloop and want to put your radiators and GPU/board blocks into the other subloop, you'll get the same water delta performance, but you'll get higher flow where it matters (at the CPU block).
Where this won't work as well is with a VERY restrictive CPU block and a lack of other components in the loop. You have a good chance of actually lowering the flowrate through the block (where it matters most).
Cliffnotes version: running in parallel allows for a reduced pressure drop of the loop overall, because of the nature of the D5, this reduced pressure drop allows the pump to have increased flowrate through the pump. That increase in flowrate actually allows the pump to run more efficiently, leading to the pump actually being more "powerful" (from the liquid's perspective) and increases flowrates in equilibrium even further. The net result is that a well planned and balanced set of subloops will actually allow you to have HIGHER flowrates through each subloop (or a specific subloop) than you would if you arranged it in one serial loop.
Because the efficiency curve of a DDC peaks at lower GPM (and is narrower), I don't see this working with a DDC as effectively. Also note that this parallelization won't be effective on all loops, but there are certainly cases where it'd work well
Bookmarks