Description of problem:
TCP throughput between 2 Standard_DC96as_v5 CVMs is lower than advertised. This is reproduced at both Red Hat and Microsoft sides. Below is a test sample from NTTTCP.

Connections=1 : throughput=2.99Gbps cyclesPerByte_sender=1.65 cyclesPerByte_receiver=2.89 Avg_TCP_lat=479.864 pktsPerInterrupt=1.89 conCreatedTime=633 retransSegs=1214
Connections=2 : throughput=6.10Gbps cyclesPerByte_sender=1.78 cyclesPerByte_receiver=2.89 Avg_TCP_lat=229.669 pktsPerInterrupt=1.88 conCreatedTime=675 retransSegs=1992
Connections=4 : throughput=11.15Gbps cyclesPerByte_sender=1.48 cyclesPerByte_receiver=2.98 Avg_TCP_lat=6963.867 pktsPerInterrupt=2.42 conCreatedTime=1261 retransSegs=3889
Connections=8 : throughput=15.04Gbps cyclesPerByte_sender=1.60 cyclesPerByte_receiver=2.74 Avg_TCP_lat=7720.227 pktsPerInterrupt=3.08 conCreatedTime=1036 retransSegs=9885
Connections=16 : throughput=16.80Gbps cyclesPerByte_sender=1.81 cyclesPerByte_receiver=3.03 Avg_TCP_lat=7712.378 pktsPerInterrupt=2.43 conCreatedTime=1811 retransSegs=18803
Connections=32 : throughput=16.20Gbps cyclesPerByte_sender=2.10 cyclesPerByte_receiver=3.70 Avg_TCP_lat=8525.040 pktsPerInterrupt=2.00 conCreatedTime=3122 retransSegs=57473
Connections=64 : throughput=16.05Gbps cyclesPerByte_sender=2.93 cyclesPerByte_receiver=4.61 Avg_TCP_lat=8990.453 pktsPerInterrupt=2.33 conCreatedTime=14756 retransSegs=211787
Connections=128 : throughput=15.46Gbps cyclesPerByte_sender=4.09 cyclesPerByte_receiver=6.06 Avg_TCP_lat=9951.902 pktsPerInterrupt=3.34 conCreatedTime=30995 retransSegs=558881
Connections=256 : throughput=14.36Gbps cyclesPerByte_sender=5.69 cyclesPerByte_receiver=9.44 Avg_TCP_lat=11873.313 pktsPerInterrupt=5.48 conCreatedTime=51052 retransSegs=1118302
Connections=512 : throughput=12.84Gbps cyclesPerByte_sender=6.92 cyclesPerByte_receiver=11.98 Avg_TCP_lat=17647.435 pktsPerInterrupt=7.60 conCreatedTime=58597 retransSegs=11032245
Connections=1024 : throughput=12.91Gbps cyclesPerByte_sender=12.05 cyclesPerByte_receiver=12.39 Avg_TCP_lat=13498.971 pktsPerInterrupt=44.58 conCreatedTime=99941 retransSegs=2622917

Also notice that of Standard_DC64as_v5 is below advertised too:
Connections=1 : throughput=2.96Gbps cyclesPerByte_sender=1.82 cyclesPerByte_receiver=2.72 Avg_TCP_lat=257.609 pktsPerInterrupt=1.68 conCreatedTime=1188 retransSegs=1275
Connections=2 : throughput=5.64Gbps cyclesPerByte_sender=1.52 cyclesPerByte_receiver=2.90 Avg_TCP_lat=262.646 pktsPerInterrupt=2.11 conCreatedTime=1240 retransSegs=2265
Connections=4 : throughput=7.95Gbps cyclesPerByte_sender=1.26 cyclesPerByte_receiver=2.33 Avg_TCP_lat=7391.646 pktsPerInterrupt=2.28 conCreatedTime=1432 retransSegs=4785
Connections=8 : throughput=12.11Gbps cyclesPerByte_sender=1.50 cyclesPerByte_receiver=2.89 Avg_TCP_lat=597.900 pktsPerInterrupt=2.21 conCreatedTime=1441 retransSegs=17272
Connections=16 : throughput=12.23Gbps cyclesPerByte_sender=1.57 cyclesPerByte_receiver=3.45 Avg_TCP_lat=9136.253 pktsPerInterrupt=1.97 conCreatedTime=1957 retransSegs=30081
Connections=32 : throughput=15.70Gbps cyclesPerByte_sender=1.85 cyclesPerByte_receiver=3.75 Avg_TCP_lat=9050.949 pktsPerInterrupt=2.01 conCreatedTime=3195 retransSegs=64960
Connections=64 : throughput=15.53Gbps cyclesPerByte_sender=2.35 cyclesPerByte_receiver=4.61 Avg_TCP_lat=9142.252 pktsPerInterrupt=2.36 conCreatedTime=11085 retransSegs=202511
Connections=128 : throughput=14.88Gbps cyclesPerByte_sender=3.24 cyclesPerByte_receiver=6.16 Avg_TCP_lat=10435.140 pktsPerInterrupt=3.27 conCreatedTime=25293 retransSegs=548462
Connections=256 : throughput=13.64Gbps cyclesPerByte_sender=4.50 cyclesPerByte_receiver=9.63 Avg_TCP_lat=13677.645 pktsPerInterrupt=5.14 conCreatedTime=53623 retransSegs=1105525
Connections=512 : throughput=12.05Gbps cyclesPerByte_sender=5.30 cyclesPerByte_receiver=11.59 Avg_TCP_lat=17504.668 pktsPerInterrupt=6.28 conCreatedTime=72665 retransSegs=13815629
Connections=1024 : throughput=10.09Gbps cyclesPerByte_sender=7.65 cyclesPerByte_receiver=18.13 Avg_TCP_lat=29683.466 pktsPerInterrupt=8.21 conCreatedTime=98030 retransSegs=16183693

Version-Release number of selected component (if applicable):
5.14.0-160.el9.x86_64

How reproducible:
100%

Steps to Reproduce:
Create 2 Standard_DC96as_v5 CVMs on same subnet and use iperf3 or NTTTCP to test throughput between them.

Actual results:

Expected results:
Standard_DC64as_v5 20000 Mbps
Standard_DC96as_v5 30000 Mbps

Additional info:
Original thread - https://bugzilla.redhat.com/show_bug.cgi?id=2124751