Known regressions with recent kernels and QEMU/KVM guest networking speed?

Hi there, are there any known issues with with recent kernels versions and QEMU/KVM guest networking speed? I run two VPS instances with a cloud provider that uses QEMU and KVM, lets call them VPS1 and VPS2. I was trying to transfer a single file about 3GB in size between VPS1 and VPS2, which are located in different geographical regions, using scp or rsync. One is running openSUSE MicroOS with a 6.x kernel, and the other one is running Clear Linux also with the latest 6.x kernel. I also transferred a 3GB file from a different VPS provider (let’s call it EXT1) into VPS1 and VPS2 and experienced the same issue. The first few seconds of the transfer are very fast at ~100MB/s, but it quickly drops to around 300KB/s. I ran an mtr trace between the two instances and there is no packet loss. So then I tried the EXT1 instance with the other provider to transfer in the same file, but this time it was into a Debian 11 VPS (let’s call it VPS3 with the 5.10 kernel, and the transfer speed was very good. So that made me suspect that something was wrong with the two VPSes that are running the newer 6.x kernel. So on a whim I enabled TCP BBR congestion control on both VPS1 and VPS2, and the problem was instantly and reproducibly solved. So why is network performance good on the old Debian kernel without any sysctl tweaks (and presumably no congestion control), whereas network performance is awful with the 6.x kernel and no tweaks but good with TCP BBR congestion control enabled?