Traffic drops when using flow offload for nftables based NAT

[tech <tech@xxxxxxxxxxxxx>]

Hi,
I'm trying to augment my nft based NAT server with flow offload feature.
Prerequisites:
# uname -a
Linux nat40g 5.4.0-66-generic #74-Ubuntu SMP Wed Jan 27 22:54:38 UTC 
2021 x86_64 x86_64 x86_64 GNU/Linux
ethtool -G enp1s0f0 tx 8192
ethtool -G enp1s0f0 rx 8192
ethtool -K enp1s0f0 hw-tc-offload on

Ethernet controller: Mellanox Technologies MT27700 Family [ConnectX-4]

# cat /opt/nftables.conf
flush ruleset

table ip filter {
        chain input {
                type filter hook input priority 0; policy accept;
                ct state established accept
                iif "vlan4" counter drop
                iif "vlan5" counter drop
        }

        flowtable fastnat {
                hook ingress priority 0
                devices = { vlan4, vlan5 }
        }

        chain forward {
                type filter hook forward priority 0; policy accept;
                ip protocol { tcp , udp } flow offload @fastnat;
        }
}


table ip nat {
        chain post {
                type nat hook postrouting priority 100; policy accept;
                ip saddr 10.0.0.0/8 oif "vlan5" snat to 
19.2.5.1-19.2.5.125 persistent
        }

        chain pre {
                type nat hook prerouting priority -100; policy accept;
        }
}

All good when there is up to 12G overall traffic volume, but when my 
traffic overall volume exceeds 12+G I experience input drops.
If I comment out this portion of configuration:
flowtable fastnat {
        hook ingress priority 0
        devices = { vlan4, vlan5 }
}

chain forward {
        type filter hook forward priority 0; policy accept;
        ip protocol { tcp , udp } flow offload @fastnat;
}
The result no drops up to 21.5G and occur when CPU utilization up to 
85%.
P.S. If someone interested I can share images.