Like the NETGEAR R8000, the AC87 has a multiprocessor architecture. But the Quantenna QSR1000's embedded Synopsys ARC 700 32 bit RISC processor handles only 5 GHz wireless radio duties. The Broadcom BCM4360 2.4 GHz radio's chores are handled by the main Broadcom BCM4709 main processor.
RMerlin confirmed that all 5 GHz band processing, including Ethernet / Wi-Fi packet conversion is handled by the Quantenna section. He also pointed me to this recent ASUS PCDIY post, extolling the AC87's superior design. The detailed block diagram from that post provided some interesting information.
Note that the 2.4 GHz Broadcom radio is connected via PCIe. Since the BCM4360 has a PCIe 2.0 interface, it has 500 MB/s (4000 Mbps) to itself in each direction. In contrast, the Quantenna QT3840 SoC connects via an RGMII interface, which is a reduced pin version of Gigabit Ethernet. (In the NETGEAR R8000, all three radios run independently of the main BCM4709 SoC and connect via a single PCIe 2.0 bus.)
The RGMII connection means the 5 GHz radio's maximum 1733 Mbps raw PHY rate is ultimately limited by a Gigabit connection back to the main processor, which handles WAN / LAN routing. And since the Quantenna-dedicated port also connects via RGMII, it too is limited to Gigabit speeds.
ASUS RT-AC87 detailed block diagram
Even more interesting is that the Quantenna SoC has its own directly-connected Gigabit Ethernet port. So Ethernet traffic to port 1 (if my interpretation of the board photo supplied by RMerlin and shown in Part 1 is correct), has to travel through two RGMII ports to get to the WAN or ports 2-4.
In the Part 1 Inside section I also missed the fact that ports 1 and 2 are marked as "Teaming port". But there are no teaming / aggregation controls to be found on any of the AC87's admin pages. Link aggregation helps for only multiple connections (slide 7) anyway. It doesn't provide a higher bandwidth "pipe" for a single connection.
Internal architecture differences notwithstanding, I subjected the AC87 to the same stress tests run on the NETGEAR R8000. Except that instead of connecting each of the two 5 GHz clients to its own radio, both were connected to the single Quantenna-based radio. As in the R8000 test, a separate client was connected to the 2.4 GHz radio. Also as in the R8000 stress test, one of the computers connected to the AC87's LAN ports streamed traffic to both the single 2.4 GHz client and one of the 5 GHz.
ASUS RT-AC87 Stress Test Setup
I first ran IxChariot throughput.scr scripts using TCP/IP and test file sizes of 9,000,000 Bytes for 5 GHz and 5,000,000 Bytes for 2.4 GHz. Each radio was hit with a single IxChariot stream. There was no wired routing traffic in this test.
The screenshot below is a composite of the three wireless connections run individually to establish a baseline. Total throughput here is 1260.5 Mbps. This result would have been higher if I could subtract out the low throughput at the start of the 2.4 and 5 GHz runs caused by an IxChariot quirk. The large throughput swings in the 5#1 plot line are not IxChariot quirks; similar disturbances were found during many 5 GHz runs.
ASUS RT-AC87 Wireless Stress Test - Individual run composite
The next screenshot shows all three streams run simultaneously. 2.4 GHz radio throughput is about the same. The two 5 GHz streams appear to produce about 700 Mbps of total throughput vs. the 550 Mbps or so produced by each single stream. This is just shy of a 30% gain. Remember, the 5 GHz radios are three-stream, and can't access the higher bandwidth possible with a four-stream client.
ASUS RT-AC87 Wireless Stress Test - Simultaneous runs
By comparison, the dual 5 GHz radio architecture of the NETGEAR R8000 produces a total of over 1000 Mbps throughput (750 + 275 via eyeballing traces), shown in the plot below taken from the R8000 Part 2 review. Throughput variation for each 5 GHz client is also much less than for the single-radio AC87 design.
For highest 5 GHz throughput from multiple clients, it's clear that the multi-radio XStream architecture in the NETGEAR R8000 wins.