The following is the `perf top` output of blue's CPU2 (which has been made to handle both tb0 and eno1) when running iperf3 from red->purple (the slow, ~5Mbps direction).

   PerfTop:     788 irqs/sec  kernel:94.7%  exact: 100.0% lost: 0/0 drop: 0/0 [4000Hz cycles:P],  (all, CPU: 2)
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

     3.23%  [kernel]       [k] e1000_irq_enable
     2.99%  [kernel]       [k] ioread32
     2.94%  [kernel]       [k] cpuidle_enter_state
     2.56%  [kernel]       [k] e1000_intr_msi
     2.14%  [kernel]       [k] menu_select
     1.93%  [kernel]       [k] _raw_spin_lock
     1.78%  [kernel]       [k] clflush_cache_range
     1.43%  [kernel]       [k] _raw_spin_lock_irqsave
     1.41%  perf           [.] dso__find_symbol
     1.39%  [kernel]       [k] pfn_to_dma_pte
     1.18%  [kernel]       [k] clear_page_erms
     1.00%  [kernel]       [k] br_handle_frame
     0.97%  [kernel]       [k] __dev_queue_xmit
     0.89%  [kernel]       [k] intel_iommu_map_pages
     0.82%  [kernel]       [k] native_irq_return_iret
     0.76%  [kernel]       [k] dma_pte_clear_level
     0.73%  [kernel]       [k] do_idle
     0.71%  [kernel]       [k] alloc_iova_fast
     0.71%  [kernel]       [k] __hrtimer_next_event_base
     0.69%  [kernel]       [k] __schedule
     0.69%  [kernel]       [k] read_tsc
     0.67%  [kernel]       [k] __netif_receive_skb_core.constprop.0
     0.67%  [kernel]       [k] native_apic_msr_eoi
     0.67%  [kernel]       [k] timerqueue_add
     0.66%  [kernel]       [k] psi_group_change
     0.66%  [kernel]       [k] native_sched_clock
     0.65%  [kernel]       [k] csum_partial
     0.63%  [kernel]       [k] net_rx_action
     0.59%  [kernel]       [k] kmem_cache_free
     0.55%  [kernel]       [k] e1000_xmit_frame
     0.55%  [kernel]       [k] sched_balance_update_blocked_averages
     0.54%  [kernel]       [k] ktime_get
     0.52%  [kernel]       [k] __get_next_timer_interrupt
     0.51%  [kernel]       [k] irq_chip_ack_parent
     0.50%  [kernel]       [k] __handle_irq_event_percpu
     0.48%  [kernel]       [k] ring_msix
     0.47%  [kernel]       [k] fdb_find_rcu
     0.46%  [kernel]       [k] e1000_clean_rx_irq
     0.46%  [kernel]       [k] eth_type_trans
     0.45%  [kernel]       [k] tbnet_poll
     0.41%  [kernel]       [k] rcu_sched_clock_irq
     0.41%  [kernel]       [k] e1000_clean_tx_irq
     0.39%  [kernel]       [k] sched_balance_newidle
     0.39%  [kernel]       [k] inet_gro_receive
     0.38%  [kernel]       [k] native_write_msr
     0.38%  [kernel]       [k] tbnet_alloc_rx_buffers
     0.38%  [kernel]       [k] __switch_to_asm