We are using E1000 and VMXNET3, but this does not make a notable difference. It seams that VMXNET3 is slightly faster in both directions, but sending is still significantly slower than receiving. Here's a result-screen of a test we've made.
NETIO - Network Throughput Benchmark, Version 1.26
(C) 1997-2005 Kai Uwe Rommel
TCP connection established.
Packet size 1k bytes: 65316 KByte/s Tx, 17394 KByte/s Rx.
Packet size 2k bytes: 62178 KByte/s Tx, 19588 KByte/s Rx.
Packet size 4k bytes: 63279 KByte/s Tx, 19111 KByte/s Rx.
Packet size 8k bytes: 77341 KByte/s Tx, 22257 KByte/s Rx.
Packet size 16k bytes: 80261 KByte/s Tx, 37704 KByte/s Rx.
Packet size 32k bytes: 89565 KByte/s Tx, 70087 KByte/s Rx.
Done.